Jonathan Gratch

1
The Architectural Role of Emotion in Cognitive
Systems (is emotion Type I or Type II?)

• Jonathan Gratch
• USC Institute for Creative Technologies
• Joint work with
• Stacy Marsella
• USC Information Sciences Institute
• Sponsored by the U.S. Army Research,
  Development, and Engineering Command. The
  contents are those of the author and do not
  necessarily reflect the views of the Department
  of the Army

2
Outline

• Human emotion can inform intelligent system
  design
• Inform outer-directed behavior
• Human Computer Interaction
• Inform inner-directed behavior
• Decision making of the intelligent system
• Ground discussion in application
• Mission Rehearsal Exercise system
• Appraisal Theory
• Illustrate impact on architecture design
• Experimental results

3
Emotion as a External Behavior

• Users readily read expressive behaviors
• Communicates information
• Can influence the interaction
• Evolutions solution to prisoner dilemma? Frank

4
Emotion as Mental Behavior

• Emotion influences mental functions
• Maintaining Homeostasis
• Focus of attention
• Goal prioritization
• Learning
• Belief formation
• Decision-making
• Preparing action and reaction
• Can emotion research inform solutions for
  artificial minds

5
Grounding Virtual Humans

• Interactive
• Perceive the world
• Think for themselves
• Speak
• Integrated architecture
• Perception, Planning, Action
• Natural Language Processing
• Non-verbal communication
• Teamwork
• Emotion

Psychotherapy
Training
6
Mission Rehearsal Exercise

• Social Training Simulation
• Explore high-stakes social interactions in safety
  of VR

Rickel, Marsella, Traum, Hill, Hovy, Johnson,
Narayanan, Swartout,
7
MRE Leadership Training
8
SASO Negotiation Trainer
9
Cognitive Architecture Soar
Future
Present
Past
Eagle 1-6 Assist Desire LT Belief False
Child HealthyFalse
Accident Intend False Blame unresolved
Assist Eagle 1-6False
Get Medevac ResponsibilityLT Intend True
Medevac AvailableTrue
Child-Healthy Desire SGT Belief
False Probability 75
Working Memory
Planning
Perception
Dialogue
Action
Soar operators
10
Cognitive Representation
Future Plans
Present
Past Events
Eagle 1-6 Assist Desire LT(50) Belief False
Child HealthyFalse
Accident Intend False Blame unresolved
Assist Eagle 1-6False
Get Medevac ResponsibilityLT Intend True
Medevac AvailableTrue
Child-Healthy Desire SGT(80) Belief
False Probability 75

• Causal Interpretation
• Agent centric subjective view
• Uniform representation of past, present, future
• Combines decision-theoretic plans with models of
  belief and intention

11
Architectural Role of Emotion(A trip from Type
II to Type I)

• Began with view emotion as veneer
• Ended up as central organizing construct
• Initial problem
• how to convey emotion in interactive setting?
• Built mechanism to infer plausible emotions
• In response to simulation events
• In response to user interventions
• But discovered resolved architectural issues
• Coherence is more than skin deep
• Build it and they will come

12
How to convey emotion

• Mind/Body problem
• Emotions arises from cognitive assessments
• Aristotle, et. al
• Emotions arise from sub-symbolic processes
• William James, et. Al
• 2 process models
• LeDoix, Damasio, Busemeyer
• Pragmatics
• Artificial intelligence techniques largely
  symbolic
• Planning
• Explanation
• Dialogue management
• Favors cognitive theories

13
How to convey emotion

• Appraisal Theory
• Influential and well-established theory
• Arnold, Frijda Lazarus Ortony, Clore
  Collins Scherer Smith
• Emphasizes tight coupling between
• Emotion
• Cognition
• Motivation

14
Cognitive Appraisal Theory
Smith and Lazarus cognitive-motivational-emotive
system
Appraisal
Appraisal
Internal goals, beliefs
External World
Appraisal
Appraisal Variables
Physiological Response
Action Tendencies
Affect
Appraisal
Coping
Appraisal
15
Appraisal
Appraisal
Appraisal
Internal goals, beliefs
External World

• Appraisal relational assessment
• Compare beliefs, desires and intentions
• with
• external circumstances

16
Appraisal

• Characterize via 72 appraisal variables
• Desirability
• Likelihood
• Urgency
• Unexpectedness
• Causal attribution (causality, agency, coercion,
  blame)
• Coping potential (controllability, adaptability)
• Mediate behavior
• Decision and action
• Physiological changes and facial expressions
• Personality and cultural differences

17
Appraisal

• Characterize via 52 appraisal variables
• Desirability
• Likelihood
• Urgency
• Unexpectedness
• Causal attribution (causality, agency, coercion,
  blame)
• Coping potential (controllability, adaptability)
• Superset of criteria considered by intel systems
• Decision theory desirability, likelihood
• Scheduling desirability, urgency
• Requires additional Type II machinery

18
Coping Strategies

• Coping Response strategy
• Characterized by ontology of coping strategies

Internal goals, beliefs
External World
Appraisal
Coping
Problem-focused
Emotion-focused
Appraisal
19
Coping Strategies

• Problem-focused (act on the world)
• Action execution
• Planning
• Seek information
• Seek instrumental social support
• Analogous to
• Deliberative or reactive problem solving
• Teamwork

20
Coping Strategies

• Emotion-focused (act on representation)
• Avoidance
• Denial
• Shift blame
• Distancing
• Not typically considered by intelligent systems
• More than a decision (e.g. abandon current plan)
• Provides self-justification for why
• Related to motivational / explanatory coherence
• Leads to persistent change in behavior
• Requires additional Type II machinery

21
Appraisal
Appraisal
Internal goals, beliefs
External World
Appraisal
Cognition
Appraisal
Coping
Appraisal
22
EMA Model of Appraisal and Coping
Future
Past
Working Memory
Planning
Perception
Dialogue
Action
Soar Operators
23
EMA Model of Appraisal and Coping

• Appraisal as plan-evaluation
• Causal interpretation mediates agent-environment
  relationship
• Characterize features of interpretation via
  appraisal variables
• Fast, reactive, parallel (Soar elaboration
  rules)
• Coping strategies alternative Type II choices
• Problem-focused
• Planning ? execute step, add plan step,
• Seek information ? Dialogue, perception
• Emotion-focused
• Denial ? Change belief
• Find silver lining ? Change utilities
• Shift blame ? Change causal attribution
• ? Dialogue moves
• Distancing ? Drop goal / intention

24
Focus of attention

• Problem too much information in memory
• Agent awash in emotion
• Added focus of attention model
• Sub-symbolic spreading activation
• E.g. to answer What happened here, must access
  all events in causal history
• Collect appraisals associated with each event
• Pick most intense appraisal to modify behavior of
  operator

25
Causal Interpretation (Goals, Beliefs, Causal
Relations, Plans, Intentions)
Environment
Action
Planning
Dialogue
Type II Processes
Belief Formation
Perception
Explanation
26
Causal Interpretation (Goals, Beliefs, Causal
Relations, Plans, Intentions)
Environment
Appraisal
Action
Planning
Dialogue
Belief Formation
Appraisal Frames
Affective State
Perception
Explanation
Coping
Control Signals
Control Signals
27
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame unresolved
Child-Healthy Desire SGT Satisfied False
Perspective Self (Sgt) Desirability
-80 Likelihood 100 Blame/Credit unresolved
Distress 80
Sgts Appraisal of Accident from his perspective
28
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame unresolved
Child-Healthy Desire SGT Satisfied False
Distress 80
29
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame unresolved
Child-Healthy Desire SGT Satisfied False
Sgts Own Perspective
Distress 80
30
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame unresolved
Child-Healthy Desire SGT Satisfied
False Probability 75?
Get Medevac ResponsibilityLT
Make Amends
Distress 80
Distress 80
Problem-Focused Coping Form intention to help
Boy
31
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame MOM
Child-Healthy Desire SGT Satisfied
False Probability 75?
Get Medevac ResponsibilityLT
Shift Blame
Make Amends
Distress 80
Distress 80
Emotion-Focused Coping Blame Mother
32
Eagle 1-6 Assist Desire LT Satisfied False
Child HealthyFalse
Assist Eagle 1-6False
Accident Blame unresolved
Child-Healthy Desire SGT Satisfied
False Probability 75?
Get Medevac ResponsibilityLT
Shift Blame
Make Amends
Distress 80
Distress 80
Appraisal Variables
33
Emotion as a Type I system

• Appraisal
• Lingua Franca for characterizing Type II
  processes
• Coping
• imposes coherent response
• Including re-conceptualization of
  agent-environment relationship
• Functional
• Influenced the design of the integrated system
• Facilitates solutions to integration challenges

34
Solutions

• Goal prioritization/abandonment
• Linguistic reference resolution
• Integration of top-down and bottom up perception
• Social credit assignment
• Negotiation tactics
• Appraise state of a conversation
• Select tactic based on perceived desirability,
  control,

35
Cognitive Plausibility

• EMA as model of human emotion
• Given some situation, model should generate
• Right emotion but also
• Right appraisals
• Right coping strategies
• Right dynamics
• How these change as a situation evolves
• Preliminary evaluations against human data

36
Outline
Evaluate EMA with Clinical Instrument

• Profile of normal emotional responses
• Stress and Coping Process Questionnaire
• Perrez, M., Reicherts, M. (1992)
• Used to identify clinically depressed patients
• Query imagined response to synthetic stimuli
• Onset You forgot to do something important for
  your partner. They get very angry and blame you
• Continuation After a while, their attitude
  hasnt changed
• Outcome They leave in an angry state.

37
Aversive Condition
? Past
Present
Future ?
Goal Utility 100 Probability 66 Intend-that
True
Goal Utility 50 Probability 100 Belief False
Past Act Prob 100 Responsibility Other
Establishes
Clobbers
Future Act Intend-to True Probability
66 Responsibility self
Desired state facilitated Desirability
66 Likelihood 50 Attribution self Emotion
Hope(66)
Desired state threatened Desirability
-100 Likelihood 50 Attribution Other Emotion
Fear(50), Anger(50)
38
a) Controllability Loss
Aversive
HUMAN
Data fits normal profile
HUMAN
EMA
EMA
b) Changeability Loss
Aversive
EMA
EMA
HUMAN
HUMAN
c) Valence Loss
Aversive
EMA
EMA
HUMAN
HUMAN
39
Summary

• Appraisal theory productive model of emotion for
  intelligent systems
• Gives insight on type II processes required for
  intelligent behavior
• Provides coherent control of these processes
• EMA practical model of these processes
• Some fit to human data