Teaming up humans with autonomous synthetic characters Teaming up humans with autonomous synthetic...

Preview:

Citation preview

Teaming up humans with autonomous synthetic characters

소프트웨어 에이전트2009. 06. 06

이승현

R. Prada, AI Communicationsvol.21, pp.80-103 2008

S FT COMPUTING @ YONSEI UNIV . KOREA

Contents

• Introduction

• Fundamentals of Group Dynamics– Group– Interaction– Group structure

• SGD Model – Design specification

• Case study: Perfect Circle– Setting– Result

• Conclusion

2 /21

S FT COMPUTING @ YONSEI UNIV . KOREA

Introduction

• Issue on synthetic characters– Making characters believable and creating the illusion of life

• Problems– Lack of necessary social skills to interact in a group– Limited role and autonomy

• Goal– Create team-oriented autonomous agents with social skills– Create characters which are able to behave coherently with the

group’s composition, context and structure

• Theoretical basis– Human social psychological sciences

3 / 21

S FT COMPUTING @ YONSEI UNIV . KOREA

Fundamentals of Group Dynamics

• Definition of group– Interaction– Interdependency– Mutual perception and identification

• Group dynamics– System and Interaction theories[Cartwright and Zander, 1968]

4 / 21

Group

S FT COMPUTING @ YONSEI UNIV . KOREA

Fundamentals of Group Dynamics

• Interaction Process Analysis[Bales, 1950]– Two classes of problems

5 / 21

Interaction

Instrumental Problems

Active Interaction

Passive Interaction

Socio-emotional Problems

Possitive

PassiveNegative Interaction

S FT COMPUTING @ YONSEI UNIV . KOREA

Fundamentals of Group Dynamics

• Main influential factors on group structure[Jesuino, 2000]– Structure of communication– Structure of social power– Structure of social attraction

• Social Power[French & Raven, 1968]

: Potential influence exerted by a social agent on a person

6 / 21

Group Structure

Category description

Reward power based on perceived ability to mediate rewards

Coercive power based on the perceived ability to mediate punishments

Legitimate power based on the perception that someone hasthe right to prescribe given behaviours

Referent power based on perceived associations between the person and the social agent

Expert power based on the perceived distinctive knowledge, expertness, abilities or skills

Information power based on the perceived control of the information needed

S FT COMPUTING @ YONSEI UNIV . KOREA

Fundamentals of Group Dynamics

• Social attraction: affective attitude of each member of the group towards the other

members [Moreno, 1934]– Balance Theory [Heider, 1946]

: people have to maintain balanced cognitive configurations• Concept of a POX triple which represents a cognitive configuration

7 / 21

Group Structure

Stable POX cognitive configurations

Unstable POX cognitive configurations

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Synthetic Group Dynamics(SGD) Model– Principle

• Must be aware of the other members and the group itself• Able to build proper knowledge regarding group’s social structure • Able to use the knowledge to drive its behavior

– Target groups• Group = Human user + several autonomous synthetic characters• Small groups with only a few members• Groups without a strong organizational structure

8 / 21

Principle and Target

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Structure of SGD model(overview)

9 / 21

Overview

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Design of individual– Agent’s abilities

• Levels of expertise of which agent can perform• Used when define agent’s position in a group

– Agent’s personality(Five Factor Model, 1996)• Extraversion

: Dominant initiative of the agent

: Influence on the agent’s frequency of interaction• Agreeableness

: Socio-emotional orientation of the agent

ex) Agreeable agent, less agreeable agents

• Knowledge base

10 / 21

Individual Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Design of group – Group identity: name– Composition: set of individuals that are associated with the group– Structure(Based on Jesuino’s)

• Social attraction: interpersonal attraction of the members

ex) like, dislike• Social influence: relation of power

ex) power one can exert, power one is able to resist

• Knowledge base– Group-individual relation

11 / 21

Group Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Design of interaction

• Predicate and function(Knowledge base)

12 / 21

Interaction Level

Action Type or pattern of interaction

Performers Agents that are engaged in the execution of the interaction

Supporters Agents that support the interaction without being directly involved

Targets Agents that are affected by the interaction

Strength Relative importance of the interaction

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Classification of interaction(Based on Bale’s IPA system)

13 / 21

Interaction Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Dynamics of the interaction<Occurrence of an interaction in the group>– Agent’s motivation, group position, personality Frequency

– Agent’s personalities Tendencies of social emotional interactions

– Level of expertise Instrumental interactions

– Agent’s position Socio-emotional interactions

14 / 21

Interaction Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model(Continued)

<Instrumental interaction structure of social influence>

15 / 21

Interaction Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model(Continued)

<Socio-emotional interaction structure of social attraction>

16 / 21

Interaction Level

S FT COMPUTING @ YONSEI UNIV . KOREA

SGD Model

• Behavior generation(Algorithm)

17 / 21

Behavior Generation

S FT COMPUTING @ YONSEI UNIV . KOREA

Case Study: Perfect Circle

• Perfect Circle: the Quest for the Rainbow Pearl– Collaborative game– 4 autonomous characters and 1 user-controlled characters

• Game mechanism(overview)

• Design criterion[Mennecke and Wheeler, 2004]– It is appropriate for all the subject– It promotes subject’s intellectual engagement– It has a good level of complexity– It is conjunctive– The differences in subject’s experiences

18 / 21

Implementation

AbilityGemstonePortalMagic Item

S FT COMPUTING @ YONSEI UNIV . KOREA

Case Study: Perfect Circle

• How SGD model is applied?– 2 instrumental interactions

• Facilitate problem: interaction corresponds to the event of a successful execution of a manipulation on a gemstone

• Obstruct problem: corresponds to a failure– 4 socio-emotional interactions

• Agree/ Disagree/ Encourage/ Discourage• Intension is presented by the form of senteces

ex)”I believe you”(positive), “Stop doing that”(negative)

19 / 21

Implementation

S FT COMPUTING @ YONSEI UNIV . KOREA

Case Study: Perfect Circle

• Subjects:– 24 students of computer science(20 male , 4 females students)

• Independent variables– 2 version of game(SGD model applied or not)– Two initial group structures(neutral/hostile)

• Dependent variables– Group trust/ Group identification/ Satisfaction

• Measures– Questionnaires (modified Allen’s version)– 3 positive and 3 negative questions(7 level scale)– Reliability test: Cronbach’s Alpha value

• Procedure– 2 Hours(30 min for learning, 1 hour for playing, 30 min for answering)– 3 groups(without SGD, SGD with neural relation, SGD with hostile rela-

tion)

20 / 21

Experiments

S FT COMPUTING @ YONSEI UNIV . KOREA

Case Study: Perfect Circle

• Result(Mann-Whitney test)– Synthetic characters with SGD mode high level of trust

– Low level of initial cohesion of the group(with SGD model)

Level of trust in a group is much higher

Better identification of themselves

Subjects were more satisfied

21 / 21

Result

S FT COMPUTING @ YONSEI UNIV . KOREA

Conclusion

• Former collaboration games has lack of social interactions

• It suggests a way of making synthetic characters smarter based on some human social psychological theories

– Concept of SGD model– Implementation: Perfect Circle

• Additional result – Low level of initial group cohesion can cause more user’s satisfac-

tion

22 / 21

End of Document

Recommended