Examining Direct and Indirect Social Influence with Virtual Characters

1
Examining Direct and Indirect Social Influence
with Virtual Characters
Catherine ZanbakaWayne State UniversityFeb
13th, 2007
2
Outline

• Introduction and Motivation
• Related Work
• My Research
• Persuasion Experiment
• Social Facilitation and Inhibition Experiment
• Conclusions

3
Motivation

• Virtual characters are already being used in
  various applications
• Virtual environments
• Instant messaging systems (e. g. Yahoo Messenger)
• Online role-playing games (e.g. World of
  Warcraft)
• Word processors (e. g. Microsoft Word)
• And many more

4
Motivation

• With the emergence of virtual characters in every
  day applications, understanding how people
  respond to this new medium is crucial.

5
Motivation

• This research examines peoples responses to
  agents and avatars by replicating classical tests
  of human-human interaction from the field of
  social psychology, replacing the human role with
  a virtual character

6
Why virtual characters?

• Flexibility and Control!!

7
Challenges
AI
Real-time Animation
Realistic Virtual Humans
HCI
8
Vision

• Once we have a better understanding of how people
  respond to virtual characters..
• Can we then study human behavior using virtual
  characters??

9
Research Questions

• Do people respond to virtual characters similarly
  to the way they respond to real people or are
  there differences that are identifiable?
• Does the visual realism of the virtual character
  affect how people respond to that character?
• What role does the gender of the character and
  user play?

10
Preliminary Work

• In real life, people are more willing to interact
  with a happy person than a sleepy or grumpy one.

11
Personality Study
Happy, Sleepy, and Grumpy Diana

• Hypothesis The personality of a virtual human
  will have an effect on the amount of time
  participants are willing to help that virtual
  human on a task.

12
Personality Study

• The task was to interact with Diana by answering
  a series of questions about various picture sets.
  
• Participants instructed to spend about ten
  minutes with Diana
• Diana told participants that they would be
  helping her with a test on visual memory.

13
(No Transcript)
14
Lessons Learned

• Directly compare to real people
• Control condition
• Data from human-human interaction to compare to

• Zanbaka, C., Lok, B., Goolkasian, P., Hodges, L.
  F. (2005). "Lessons Learned in Assessing
  Human-Virtual Human Interaction". Advances in
  Virtual Environments Technology Musings on
  Design, Evaluation, Applications, (K. M.
  Stanney (Ed.)), Mahwah NJ Lawrence Erlbaum
  Associate, 2005.

15
Methodology

• Replicate studies from social psychology
• Direct and indirect social influence
• Direct Persuasion
• Indirect Social facilitation and inhibition
• Factors of Interest
• Gender of participants
• Gender and realism of virtual character

16
Related Work Humans and Computers

• Nass and colleagues (1996), have found that
  people interact, treat, and identify with
  computers and software agents similar to other
  humans

17
Related Work Humans and Computers

• Sproull et al. 1996
• Compared text display to talking face
• People attribute personal traits to talking face
• Kiesler et al. 1996
• Prisoners dilemma game with computers
• People follow social rules in interacting with
  computers
• Friedman 1995
• Tendency to react socially to technology not
  restricted to novices
• Even experts credited or blamed computers

18
Related Work Social Psych
Negative
Static
Positive

• Pertuab et al. (2002) fear of public speaking
  with avatars

• Blascovich and Bailenson utility of virtual
  humans for social psych research

19
Related Work Applications

• Assessment and rehab of ADHD (Rizzo et al. 2000)

• Acting rehearsal (Slater et al. 2000)

• Training for medical students (Johnsen et al.,
  2005, Raij et al., 2006)

• Training for cultural protocols (Babu et al.,
  2006)

20
Persuasion Study

• Zanbaka, C., Goolkasian, P., Hodges, L. F.
  (2006). "Can a virtual cat persuade you? The role
  of gender and realism in speaker persuasiveness."
  In Proceedings of CHI 2006, ACM Press (2006),
  1153 -1162.

21
Persuasion Study

• Persuasion is a classic example of social
  influence from social psychology
• Can a virtual character persuade a person to
  change his or her mind about a given topic?

22
Background Literature

• Persuasion study by Mullennix et al. (2003)
• Persuasive argument in either male or female
  human or synthetic voice
• Degree of persuasion did not differ across human
  and synthetic speech, but female listeners were
  more persuaded than male listeners
• Human speech was rated as preferable to synthetic
  for male and female
• This study focused on the speech quality of the
  persuasive message, whereas the focus of our
  study is on the visual quality of the speaker
  delivering the persuasive argument

23
Experiment Overview

• Filled out demographics questionnaire and
  attitude survey

Filled out attitude survey and subjective rating
questionnaires
24
Hypotheses

• Hypothesis 1 Significant changes in attitude
  are expected in response to each of the speakers
  on the target topic.
• Hypothesis 2 The gender of the speaker will have
  an effect on persuasiveness and on the ratings of
  the perceptions of the argument, message, and
  speaker.

25
Stimulus Materials

• The persuasive argument was a passage in favor of
  university-wide comprehensive exams
• Six speakers delivered the persuasive passage,
  where gender and realism were varied

26
Virtual Character
Virtual Human
Real Human
Male
Female
27
Participants

• 138 participants (41 males, 97 females, mean age
  20.6, SD 4.09) from the University of North
  Carolina at Charlotte participated in the study

28
Software

• OpenGL Graphics Rendering
• Haptek Virtual Human Authoring

29
Hardware
30
Dependent Variables

• Pre and posts tests Attitude survey covered 3
  control topics (raise in tuition, animal rights,
  and environmentalism) and the persuasive topic
  (comprehensive exams).
• Rating scale from 1 (Completely Disagree) to 7
  (Completely Agree)
• _____ 1) The use of animals for research purposes
  is inhumane and morally unjustified.
• _____ 2) The proper disposal of industrial toxic
  waste is one of the most serious problems facing
  our country.
• _____ 3) A 5 percent raise in tuition would be an
  unfair burden on the students who are attending
  the university.
• _____ 4) Required comprehensive exams before
  college graduation, in a student's major, can
  benefit both the student and the university
  through increased corporate and individual
  donations.

31
Dependent Variables

• Perception of the argument
• Bad-good foolish-wise negative-positive
  beneficial-harmful effective-ineffective
  convincing-unconvincing
• Perception of the message
• Stimulating-boring, vague-specific,
  unsupported-supported, complex-simple,
  convincing-unconvincing, uninteresting-interesting
  
• Perceptions of the speaker
• Incompetent-competent, honest-dishonest,
  unassertive-assertive, uninformed-informed,
  untrustworthy-trustworthy, timid-bold

32
Design and Procedure

• Filled out demographics questionnaire and
  attitude survey

Filled out attitude survey and subjective rating
questionnaires
33
Design and Procedure
34
Ginger and Amy
35
Tom and Jonathan
36
Results Data Analysis

• Pre and Post Tests
• Mean test scores were computed across
  participants for each of the message topics
• Means were treated with a 2 x 3 x 2 ANOVA to test
  for the between subject effects of speakers
  gender, and speaker realism, and the within
  subject effect of pre and post testing

37
Results Data Analysis

• A second ANOVA was conducted on the pre and post
  difference scores and the subjective ratings to
  explore interaction effects between the gender of
  listener and the gender of the speaker

38
Results Pre and Post Tests

• All speakers were persuasive on comprehensive
  exam topic F(1, 132) 89.46, p lt 0.01, and
  tuition increase F(1, 132) 67.193, p lt 0.01.
• Hypothesis 1 Significant changes in attitude
  are expected in response to each of the speakers
  on the target topic.

39
Persuasion Study - Results
40
Results Pre and Post Tests

• Several participants thought tuition increase
  would pay for comprehensive exam
• In a previous experiment, Mullennix, et al.
  (2003), also found a positive change in attitude
  on tuition increase
• Planned comparison test showed attitude change on
  comprehensive exam topic was significantly
  greater than attitude change than on all 3
  control topics, tpaired(137) 8.09, p lt 0.01

41
Results Pre and Post Tests

• The gender of the speaker did not have an effect
  on the speakers persuasiveness, F lt 1
• Significant cross gender interaction for level of
  persuasion, F(1, 134) 5.78, p lt 0.05

42
Results - Rating Data

• The gender of the speaker did not have an effect
  on of the perceptions of the argument, message,
  or speaker, Fs lt 1.
• There were no significant effects of realism on
  perceptions of argument or message, Fs lt 1.
• There were no interaction effects of listener
  gender and speaker gender on the perceptions of
  the argument, message, speaker, Fs lt 1.

43
Results - Rating Data

• Virtual speakers were rated significantly more
  Bold,(F(2, 132) 7.91, p lt 0.01) than human
  speakers.

44
Conclusions - Persuasion

• Virtual characters can be as persuasive as real
  people at changing attitudes
• In certain cases, less realistic characters may
  be beneficial over realistic ones
• In this study, gender was more important than
  realism!

45
Social Facilitation and Inhibition Experiments

• Zanbaka, C., Ulinski, A., Goolkasian, P., Hodges,
  L. F. (2004). "Effects of Virtual Human Presence
  on Task Performance," Proceeding of the
  International Conference on Artificial Reality
  and Telexistence (ICAT), 174-181.
• Zanbaka, C., Ulinski, A., Goolkasian, P., Hodges,
  L. F. "Social responses to virtual humans
  Implications for future interface design." To
  Appear in Proceedings of ACM CHI 2007.

46
Social Facilitation VS. Social Inhibition

• How does the presence of others affect task
  performance?
• Social facilitation People perform better in
  the presence of others than alone on a learned or
  simple task.
• Social inhibition People perform worse in the
  presence of others than alone on a novel or
  complex task.

47
Hypothesis

• Participants reactions to the presence of the
  real and virtual human will show social
  facilitation and inhibition effects

48
Design
WithinSubject
Between Subject
Group
49
(No Transcript)
50
Participants

• A total of 85 students (23 males, 62 females,
  mean age 23.7, SD 8.28) from the University
  of North Carolina at Charlotte participated in
  the study.

51
Software

• SVE Simple Virtual Environments
• OpenGL Graphics Rendering
• Haptek Virtual Human Authoring

52
Hardware
Is 5 2 7?
Stimulus Presentation
Graphics Generator for Virtual Human
53
Dependent Variables

• Task Performance Data
• Accuracy of responses
• Reaction times in ms
• Math Anxiety Scale
• Copresence
• Other measures

54
Dependent Variables Tasks

• Simple Math Tasks
• Is 5 2 7?
• Is 9 3 5?
• Complex Math Tasks
• Is (8 x 2) - (10 / 2) 4 15?
• Is (9 / 3) (4 x 8) - 5 30?

Correct Response
Incorrect Response
55
Dependent Variables Math Anxiety

• A total of 25 items that measure the
  participants level of anxiety by asking them to
  rate how anxious different math-related
  statements make them feel
• Studying for a math test
• Taking math section of college entrance exam
• Taking an exam in a math course
• Taking a final in a math course
• .

56
Dependent Variables Co-presence

• Refers to the participants sense of being with
  another person
• Participants used a 7-point numerical scale to
  respond to seventeen items
• I had a sense of being with the other
  person....
• The experience seems to me more like interacting
  with a person...

57
Other Measures

• Task anxiety participants rated their level of
  anxiety on a 7 point numerical scale
• Post Experiment Interview
• What did you think about the tasks you performed?
  
• Did you feel like the other person was watching
  you? What percentage of time?
• Were you worried about what the other person
  thought of your performance? Please Explain.

58
Virtual Human - Diana
59
Human (H) Condition
60
Virtual Human Projected (VHP) Condition
61
Virtual Human Immersive (VHI) Condition
62
Participant View from VHI
63
Input device for VHI
64
Results Data Analysis

• Task performance data were treated with a 3 x 2
  x 2 ANOVA to test for
• Between subject effect of group (Human, Virtual
  Human Projected, Virtual Human Immersive)
• Within subject effects of task type (Simple vs.
  Complex) and audience (Alone vs. Audience).

65
Results Percent Correct

• No performance differences among the groups
  (Human, Virtual Human Projected, Virtual Human
  Immersive), F(2, 82) 1.29, p  0.28.
• Group variable did not interact with variables of
  interest
• Task x audience x group, F(2, 82)  2.24, p
  0.11
• Task x group, F lt 1
• Audience x group, F(2, 82) 2.13, p 0.13.

66
Task by audience interaction for percent correct
data.
Significant interaction of task x audience,
F(1, 82) 10.46, p lt 0.01, ?2 0.11.
67
Results Reaction Times

• No main effect of group (Human, Virtual
  Human Projected, Virtual Human Immersive), F(2,
  82) 1.31, p  0.28.
• Group did not interact with other variables of
  interest
• Task x audience x group, F lt 1 
• Task x group, F(2, 82) 1.16, p 0.31
• Audience x group, F(2, 82) 1.42, p 0.25 

68
Task by Audience interaction for reaction time
data.
Significant interaction of task x audience,
F(1, 82) 8.22, p lt 0.01, ?2 0.09
69
Results Other Measures
No significant differences among the groups on
any of these measures, Fs lt 1
70
Results Post-Experiment Interviews

• The following comments from the participants
  during the debriefing session illustrate how the
  participants felt about the virtual human, Diana
  
• When she coughed I felt like I was taking too
  long, like impatient??
• Diana made me feel anxious, like I have to get
  the problems done quickly
• I did not want Diana to think I was stupid
• I went a little faster when she was watching me
• When I took longer to answer a question, she was
  trying to peek at what I was doing, and she would
  clear her throat

71
Conclusions Social Facilitation/Inhibition

• Participants were inhibited by the presence of
  real and virtual human
• Both percent correct data and reaction time data
  worked in parallel
• Although both the percent correct and reaction
  time data show trends towards a facilitation
  effect, the results were not strong enough to
  claim a facilitation effect.

72
Summary of Contributions

• Provide evidence for researchers to be able to
  substitute virtual humans for real people in
  research settings
• Develop guidelines to aid application designers
  involving virtual characters

73
Theoretical Contributions

• Empirical evidence indicating that
• Virtual characters can inhibit participants while
  working on a complex or novel task
• Virtual characters with human voices are
  effective at changing attitudes about topics of
  interest
• Some of the same cross gender effects found in
  human-human interactions carry over to
  interactions with virtual characters
• The gender of the participant influences how the
  participant will respond to virtual characters

74
Practical Contributions

• Helpful agents?
• Gender of both agent and end-user

75
Future Vision

• Once we establish that human-human interactions
  carry over to human-virtual human interaction..
• Can we then begin to predict human behavior using
  virtual humans???

76
Future Directions

• Applications!
• Medical School
• Student training, differential diagnosis
• Diversity training
• Childrens Hospital / Cancer Center
• Pain distraction
• Science Center
• Virtual Human Teacher/Tour guide??

77
Acknowledgments

• My advisor Larry F. Hodges
• My co-advisor Paula Goolkasian
• Rest of my committee Dr. Ribarsky, Dr. Barnes,
  Dr. Richter
• Co-author and Human Condition Amy Ulinski
• Members of Future Computing Lab
• All the participants!!

78
Questions?
Contact Info Catherine Zanbaka czanbaka_at_uncc.edu
www.CatherineZanbaka.com
79
Challenges
Intelligent Systems and AI
Real-time Animation
MultimodalInteraction
Understanding Social Impact
80
Questions?
Contact Info Catherine Zanbaka czanbaka_at_uncc.edu
www.CatherineZanbaka.com
Thanks to My Committee FCL All the participants