a formal model of social relations for artificial companions florian pecune – magalie ochs –...

27
A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

Upload: erick-hickenbottom

Post on 16-Dec-2015

217 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

A Formal Model of Social Relations for Artificial Companions

Florian PECUNE – Magalie OCHS – Catherine PELACHAUD

CNRS – LTCI, Télécom Paristech

Page 2: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

The MoCA Project

Florian Pecunepage 2

Create a “little world” of artificial companions…

… with different roles and behavior

“…a robot or a virtual conversational agent that possesses […] social skills that allow it to establish and maintain long-term relationships with users.” (Lim, 2012)

Focus on dyadic interactions

Page 3: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Thesis Objective

Florian Pecunepage 3

Create an agent able to make decisions according to its relation …

… but also to influence user’s relation

How to represent social relations?

Three research problematics:

How do they change during interactions?

I do that because I like the user

I do that because I want the user to like me

How to model the effects of these relations?

Page 4: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM Florian Pecunepage 4

Ideal Social Relation

Expressed Social Relation

Relation the agent would like to express

Intrinsic relation felt by the agent

Relation expressed by the agent

Felt Social Relation

I like the user, I am less dominant than him

I would like to be more dominant than the user

Influenced by agent’s personality (Trapnell & Wiggins, 1990)

Influenced by agent’s role (Heise, 1987)

Influenced by felt and ideal relations (Scherer, 2005)

Influenced by Mood (Isen & al., 1992)

How to represent social relations? (1/2)

Page 5: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM Florian Pecunepage 5

Ideal Social Relation

Expressed Social Relation

Relation the agent would like to express

Intrinsic relation felt by the agent

Relation expressed by the agent

Felt Social Relation

I like the user, I am less dominant than him

I would like to be more dominant than the user

Influenced by agent’s personality (Trapnell & Wiggins, 1990)

Influenced by agent’s role (Heise, 1987)

Influenced by felt and ideal relations (Scherer, 2005)

Influenced by Mood (Isen & al., 1992)

How to represent social relations? (1/2)

Page 6: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

How to represent social relations? (2/2)

Florian Pecunepage 6

Dominance

Familiarity

“the capacity of one agent to affect the behavior of another” (Prada & Paiva, 2008)

“mutual knowledge of personal information” (Svennevig, 1999)

Dominant

LikingDisliking

Submissive

Several dimensions used to define social relations Dominance and liking are the most widely used (Argyle, 1988)

Familiarity useful for long-term relationship (Bickmore & Picard, 2005)

(Altman & Taylor, 1973)

Liking

“A general and enduring positive or negative feeling about some person” (Moshkina & Arkin, 2003)

Page 7: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – The dependence (1/3)

Florian Pecunepage 7

Dependence of an agent A upon an agent B (Emerson, 1962)

“Directly proportional to A's motivational investment in goals mediated by B (…)

(…) Inversely proportional to the availability of those goals to A outside of the A-B relation”

Agent B Agent A

I really want to go to the concert tonight !

High Dependence

Agent B Agent A

I’m the only one who can bring you there.

Agent B Agent A

Maybe I could go to the concert tonight.

Low Dependence

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.We can bring

you there too !

I can bring you there with my car.

I can bring you there with my car.

I really want to go to the concert tonight !

Page 8: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – The dependence (2/3)

Florian Pecunepage 8

An agent i can be helpful ….

… or threatening

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.

)),(),((,)(),( aiAgentaFeasibleAaIjHelpful ji

Agent B Agent A

I really want to go to the concert tonight !

Not if I cancel the concert !

If j has an intention

And i can do an action a after what it will be impossible to achieve

And i can do an action a after what will be true

If j has an intention

)),())(,(()(),( aiAgentPossibleaFeasibleaIjgThreatenin ji

)),(),(),'((' BeforeaiAgentaFeasiblea

Or i can do an action a’ cancelling a precondition of

Page 9: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – The dependence (3/3)

Florian Pecunepage 9

j is dependant upon i toward an intention

i can help j to achieve this intention

j can not achieve this intention himself

i can prevent j to achieve this intention

))),(),((),((),( ajAgentaFeasibleajHelpfulBiDependant ijj

)),(( jgThreateninB ij

and or

means that j believes

Page 10: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – The dependence value (1/2)

Florian Pecunepage 10

Importance accorded by i to the success of an intention ]1;0])(_ isuccimp

]1;0])(_ ifailimp

Dependence influenced by the importance accorded to intentions…

Importance accorded by i to the non failure of an intention

… and the number of potential helpers

),()(_ iHelpfulikcardnumhelpers ki

Number of agents susceptible to have a positive impact on the intention

The number of potential threatening agents is not taken into account

Agent A

It is normal to be understood by the user

Low importance of successAgent A

It is not normal if the user do not understand me !

High importance of failure

Page 11: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – The dependence value (2/2)

Florian Pecunepage 11

Dependence value

)(_

)(_),(_

i

ii numhelpers

succimpjvaluedepend )(_),(_ ii failimpjvaluedepend

Positively influenced by the importance accorded to the intention (success or non failure)

Negatively influenced by the number of potential helpers

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.

9,0)(_ isuccimp 1)(_ inumhelpers

High Dependence

High importance Low number

Agent B Agent A

I don’t care that muchI can destroy your TV !

1,0)(_ ifailimp

Low importance

Low Dependence

Page 12: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – Dominance value

Florian Pecunepage 12

i dominance toward j Bi(j dependence upon i) Bi(i dependence upon j)

The dominance value of an agent i toward another agent j is :

The difference between his belief about j’s dependence upon him and his believed dependence upon j

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.

High DependenceAgent B Agent A

I can destroy your TV!I don’t care that much…

Low Dependence

B is dominant

The agent i is dominant toward j if he believes he is less dependant upon j than j is upon him

Page 13: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – Strategies (1/2)

Florian Pecunepage 13

To improve his level of dominance, an agent i can : Decrease the importance accorded to one of his intentions that the agent j may achieve or threaten

Decrease his dependency by finding other sources able to help him

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.

High importance Low number

Agent B Agent A

Finally, I don’t care…I can bring you there with my car.

Low number

)(_

)(_),(_

i

ii numhelpers

succimpjvaluedepend

Low importance

High number

Agent B Agent A

I really want to go to the concert tonight !

I can bring you there with my car.We can bring

you there too !

High importance

Page 14: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Dominance – Strategies (2/2)

Florian Pecunepage 14

To improve his level of dominance, an agent i can : Try to increase the motivational investment of the other agent

Deny alternative sources for the other agent

i dominance toward j Bi(j dependence upon i) Bi(i dependence upon j)

Find other intentions for which he can be helpful

Find other intentions for which he can be threatening

Page 15: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

How to represent social relations? (2/2)

Florian Pecunepage 15

Dominance

Familiarity

“the capacity of one agent to affect the behavior of another” (Prada & Paiva, 2008)

“mutual knowledge of personal information” (Svennevig, 1999)

Dominant

LikingDisliking

Submissive

Several dimensions used to define social relations Dominance and liking are the most widely used (Argyle, 1988)

Familiarity useful for long-term relationship (Bickmore & Picard, 2005)

(Altman & Taylor, 1973)

Liking

“A general and enduring positive or negative feeling about some person” (Moshkina & Arkin, 2003)

Page 16: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Liking – Theory

Florian Pecunepage 16

Based on Heider’s Balance Theory(Heider, 1958)

Concept X

Agent B Agent A

A relation is balanced :

“If all three relations are positive in all respects (…)

(….) or if two are negative and one positive.”

Concept

Agent B Agent A

Concept

Agent B Agent A

I like chocolate. Me too !

I don’t like bravery. I do.

People tend to seek balanced states (Zajonc, 1960)

Sharing appreciations will improve the liking

Page 17: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Liking – Appreciation

Florian Pecunepage 17

)(XAppreciateiThe agent i appreciate the concept X

))(()(_ XAppreciateBXAppreciateonAppreciatiShared jii

Shared appreciation

Concept

Agent B Agent A

Concept

Agent B Agent A

I like chocolate. Me too !

I don’t like bravery. I do.

))(()(_ XAppreciateBXAppreciateonAppreciatiContra jii

Contradictory appreciation

Page 18: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Liking – Liking value

Florian Pecunepage 18

i liking toward jImportance accorded to shared concepts

Importance accorded to contradictory concepts

]1;0[)(importance XiImportance accorded to a concept X

Concept

Agent B Agent A

I like chocolate. Me too !Concept

Agent B Agent A

I don’t like bravery. I do.

2.0)(importance Xi

Low liking High disliking

9.0)(importance Xi

Liking value of an agent i toward an agent j :

Page 19: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Liking – Strategies

Florian Pecunepage 19

To improve his level of liking, an agent i can : Change its own appreciation about a concept for which j has a different feeling

Try to change j’s feeling about this same concept

Concept

Agent B Agent A

I don’t like football. I do.

Unbalanced

Concept

Agent B Agent A

I like football. Me too.

Concept

Agent B Agent A

I don’t like football. Me neither.

Page 20: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

How to represent social relations? (2/2)

Florian Pecunepage 20

Dominance

Familiarity

“the capacity of one agent to affect the behavior of another” (Prada & Paiva, 2008)

“mutual knowledge of personal information” (Svennevig, 1999)

Dominant

LikingDisliking

Submissive

Several dimensions used to define social relations Dominance and liking are the most widely used (Argyle, 1988)

Familiarity useful for long-term relationship (Bickmore & Picard, 2005)

(Altman & Taylor, 1973)

Liking

“A general and enduring positive or negative feeling about some person” (Moshkina & Arkin, 2003)

Page 21: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Familiarity – Theory

Florian Pecunepage 21

FamilyW

ork

LovePolitics

Rel

igio

n

Superficial

Intimate

Personal

Core

Based on Social Penetration Theory(Altman & Taylor, 1973)

Breadth represents the number of topics shared by the agents

Depth represents the intimacy value accorded to a topic

Page 22: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Familiarity – Theory

Florian Pecunepage 22

Breadth Ratio between the number of i’s beliefs about j’s feelings and the number of possible conversational topics

N

XAppreciateBXcardjvaluebreadth

ji

i

)(()(_

]1;0[)(_ jvaluebreadthwith i Family

Work

LovePolitics

Rel

igio

n

Superficial

Intimate

Personal

Core

Agent A

5

4)(_ jvaluebreadth i

High Breadth

Except for love, I talked with j about everything I would like to know

Page 23: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Familiarity – Theory

Florian Pecunepage 23

Depth ]1;0[)( XIntimacyi

N

XAppreciateBXIntimacysumjvaluedepth

jii

i

)(()()(_

]1;0[)(_ jvaluebreadthwith i

Family

Work

LovePolitics

Rel

igio

n

Superficial

Intimate

Personal

Core

Intimacy value accorded to a concept X

5

6,1)(_ jvaluedepth i

Low Depth

Mean value of Intimacy

Page 24: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Familiarity – Strategies

Florian Pecunepage 24

To improve his level of familiarity, an agent i can : Improve breadth by asking information about topics for which he does not have any beliefs

Improve depth by asking more intimate information about a particular topic

Agent A

Let’s talk about Love !

N

XAppreciateBXcardjvaluebreadth

ji

i

)(()(_

Agent A

Tell me more about your family.

)((

)(()()(_

XAppreciateBXcard

XAppreciateBXIntimacysumjvaluedepth

ji

jii

i

Page 25: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Future Work

Florian Pecunepage 25

Define scenarios to test this model

Formalize influence of personality on social relations

Implement a neural network to represent the dynamics of the relations

Formalize ideal and expressed social relations

Page 26: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

A Formal Model of Social Relations for Artificial Companions

Florian PECUNE – Magalie OCHS – Catherine PELACHAUD

CNRS – LTCI, Télécom Paristech

Page 27: A Formal Model of Social Relations for Artificial Companions Florian PECUNE – Magalie OCHS – Catherine PELACHAUD CNRS – LTCI, Télécom Paristech

GRETA – TSI / MM

Logical Framework

Florian Pecunepage 27

Mental attitudes

)(iI

)(iB

Operators

),( aFeasible

)),(()( aFeasibleaPossible

),( Before

),( aiAgent

The agent i believes that is true

The agent i has the intention and actively desires to achieve it

The action a might happen and will be true after that

can not be true if is not already true

The agent i can be the author of the action a