The Rationality of Emotion and the emotionality of reason

1
The Rationality of Emotion and the
emotionality of reason
Eva Hudlicka Psychometrix Associates Blacksburg,
VA evahud_at_earthlink.net
MICS Symposium Saratoga Springs, NY March 4-6,
2005
2
Diseases of the Mind
Immanuel Kant
3
Outline

• Background
• Definitions Taxonomy
• Roles Effects
• Historical perspective on the how
• Cognition and Emotion
• Emotions in Integrated Cognitive Systems
• Why include them
• Requirements Architecture, Representation,
  Inferencing
• Towards a synthesis
• Modeling emotions as parameters modulating
  architecture processing
• Conclusions

4
Definition(s) of Emotions

• (See roles characteristics of emotions)
• Evaluative judgments - good / bad

5
How Do We Recognize an Emotion if we See One?

• Manifested across four modalities (Ortony,
  2001.. and others)
• Cognitive / Interpretive (Nothing is good or bad
  but thinking makes it so)
• Somatic / Physiological (neuroendocrine - e.g.,
  heart rate)
• Motivational / Behavioral (action oriented,
  expressive, visible)
• Experiential / Subjective (that special
  feeling, consciousness)
• Different emotions have different signatures
  across these modalities

6
Taxonomy Time, Differentiation, Universality

• Time
• Emotions - short (seconds, minutes) (joy, fear,
  anger)
• Moods - long (hours, days, months) (cheerful,
  depressed..)
• (Traits - very long (lifetime) (extraversion,
  neuroticism..))
• Degree of differentiation
• Affective states (non-differentiated, good / bad
  states triggering approach / avoid behaviors)
• Basic emotions (joy, sadness, fear, anger,
  disgust, surprise)
• Complex emotions (pride, guilt, shame)
• Basic emotions ----------------------gt complex
  emotions
• Triggers Universal More
  Individualized
• Behavioral manifestations Fixed Flexible
  
• Cognitive involvement Lower Higher

7
Roles of Emotions
Interpersonal
Intrapsychic
Facial expression Speech (content
properties) Gesture Posture Specific actions
8
Intrapsychic Roles of Emotions

• Perform rapid, undifferentiated processing of
  salient stimuli (avoid danger, get food)
• Trigger and prepare for fixed behavioral
  repertoires
• Rapid resource (re) allocation - mobilization
• Coordinate multiple mechanisms (attention,
  situation assessment)
• Global interrupt / alarm system signaling need
  for a shift in focus
• Interruption of on-going action
• Goal prioritization / reorganization
• Rapid resource (re) allocation - mobilization
• Motivate behavior
• Implement reward and punishment
• Enable boredom and curiosity

9
Effects on Cognition

• Anxiety and Attention WM (Williams et al.,
  1997 Mineka Sutton, 1992)
• Narrowing of attentional focus / reduction of WM
  capacity
• Predisposing towards detection of threatening
  stimuli
• Affect and Judgment Perception (Isen, 1993
  Williams et al. 97)
• Anxiety predisposes towards interpretation of
  ambiguous stimuli as threatening
• Mood biases assessment of future outcomes /
  estimates of degree of control
• Positive affect and loss aversion
• Positive affect induces global focus (forest) -
  negative affect induces local focus (trees)
• Mood and Memory (Bower, 1981 Bower, 1986)
• Mood-congruent recall
• Obsessiveness and Performance (Persons and Foa,
  1984 Sher et al., 1989)
• Delayed decision-making
• Reduced ability to recall recent activities
• Reduced confidence distinguishing btw actual
  imagined events

10
Historical Explanations of Emotions From the
Liver to Neuromodulation

• From the mortal soul (Plato) to the liver
  heart (Galen)
• From dedicated components of the brain (limbic
  system, amygdala) to specific neural circuits to
• Neuromodulation (Fellous, 2004) systemic,
  global effects across multiple brain
  structuresto
• ???

11
Neuromodulation (Fellous, 2004)
Neuromodulatory centers ( lt 10 )

• Neuromodulators (transmitters)
• Change the way in which neurons process
  information
• .Each activates distinct families of
  receptors
• Multiple time scales (msecs, hours, days)

Neurons
Emotions as dynamic patterns of neuromodulation
rather than as patterns of neural activity
(Fellous, 2004)
12
Emotion vs(?) Cognition

• Why do people get so emotional about emotion?
• Difficult to characterize investigate
• Historical focus on pathology, lack of control
  (Oh youre so emotional)
• BUT - rarely Oh youre so cognitive
• The rationality of emotion and the emotion of
  rationality
• Emotion is necessary for decision-making
• Disrupting emotional processing has impact on
  high-level cognition (Damasio)
• Search-space pruning
• Emotion AND Cognition
• Continuum of information processing
• Emotion
• Fast, undifferentiated
• Cognition
• Slower, highly differentiated

13
Why Model Emotions in Integrated Architectures?

• Snazziness and improved HCI
• Emotion recognition / generation User modeling
• Improve agent behavior
• in novel, uncertain, social environments
• by simulating what emotions do for humans
• Understand how emotions work
• Shallow vs. deep models (Sloman)

14
Narrowing Down the Problem

• Which aspects of emotions should we model?
• Cognitive / Interpretive
• Somatic
• Motivational / Behavioral
• Experiential / Subjective
• Which specific roles should we model
• Appraisal (map stimuli to resulting emotion)?
• Motivation?
• Global interrupt / alarm mechanism?
• Emotion effects on cognition?
• At what level? (deep / shallow)
• Which emotions in which contexts

15
Narrowing Down the Problem(cont.)

• Location of emotion within architecture
• Emotion as modules?
• vs. emotions as modulating parameters?
• vs. emotions as emergent properties of a complex,
  multi-level architecture?
• Particular choice depends on
• Purpose of model
• Data availability (e.g., effects on attention)
• Supporting theory (e.g., appraisal models)
• Level of fidelity desired

16
Modeling Emotion in Cognitive Architectures

• Architecture
• Levels
• Different levels support different degrees of
  affective specificity
• Functional components
• Different modules (Attention, Goal Manager,
  Affect Appraiser)
• Control data paths
• Representational and inferencing requirements
• What is represented explicitly?
• Time
• Mental constructs (situations, expectations,
  goals)
• Aspects of self environment
• What types of inferencing are supported?
• What-if (.. to generated expectations which
  influence emotions)
• Causal explanation (..important for attribution,
  regulation)
• Meta-cognition.(mediates emotion regulation,
  re-appraisal)

17
The Triune Architectures(Sloman Ortony et
al., )
Reflective / Meta-management
flexible
Full fledged affect -Basic - Complex
Complex mental models Self representations
self-awareness Explicit predictions,
causality Meta-cognition
Routine / Deliberative
Well-learned behavior Awareness Compare
alternatives - detect deviations Simple what
if Symbolic processing
Primitive Emotions Good/bad Now/later
Reactive / Reactive
Proto-affect Good/bad
Approach / Avoid Simple drives Classical
conditioning
Hardwired, fixed
18
Back to Neuromodulation
Potential for neuromodulation

• Reflective / Meta-management

HIGH
Routine / Deliberative
Reactive / Reactive
LOW
19
Towards a Synthesis Modeling Emotions as
Parameters Controlling Processing

• Emotions as patterns of modulating parameters
• Neuroscience (Fellous, 2004)
• Psychology (Matthews, 1996 Ortony, Normal,
  Revelle, 2005)
• Parameters control
• Speed, capacity, sensitivity to particular
  triggers, biases in recall and interpretation,
  latency and intensity of signals among levels
• MAMID Methodology for Analysis and Modeling of
  Individual Differences (Hudlicka, 1997 1998)
• Integrate interacting influences of multiple
  trait / state factors within single parameter
  space single simulation run

20
Demonstration Scenario

• Peacekeeping mission
• Units encounter a series of surprise events en
  route
• Unit commanders represented by an instance of
  the MAMID architecture
• Different commander types defined by distinct
  profiles
• Normal commander (E,S ), high skill level
• High anxious commander (e, s), low skill level
• High aggressive commander (E,s, A), medium
  skill level
• Commanders reactions differ
• Different commander behavior results in distinct
  mission outcomes

21
MAMID Cognitive Architecture Semantics and Data
Flow
Cues State of the world (unit attacked by
crowd)
Situations Perceived state ( unit in danger )
Expectations Expected state (unit
immobilized, casualties)
Goals Desired state (reach objective, unit
safety)
Affective state emotions Negative valence High
anxiety
Actions to accomplish goals (unit attacks
crowd)
22
Affect Appraisal
23
Emotions As Parameters
STATES /TRAITS
Processing Module Parameters (Attention /
Working Memory) Capacity Speed Inferencing
speed biases Cue selection delays
Situation selection delays ... Structural Arc
hitecture topology Weights on intermodule
links Long term memory Content structure
Attention
Situation Assessment
Traits Extraversion Stability
Conscientiousness Aggressiveness
Expectation Generator
Affect Appraiser
Affective States Anxiety Anger Sadness
Joy
Goal Manager
Action Selection
24
Modeling Threat Bias
TRAITS / STATES
Threat constructs Rated more highly
Processing Parameters Inferencing biases
Cue selection Situation selection ...

Process Threat cues
Attention
Situation Assessment
Traits Low Stability
Process Threatening interpretations
Expectation Generator
Predisposes towards
Affect Appraiser
Preferential processing of Threatening stimuli
Emotions Higher Anxiety / Fear
Goal Manager
Action Selection
25
Distinct External Events Produce Different
Anxiety Levels.. In Distinct Agent Types
Bridge / Illuminated
Hostile Crowd
Bridge
Hostile Crowd
Illuminated
26
Distinct Individual Profiles Behavior
27
Distinct Emotions Produce Different Behavior..
In Distinct Agent Types
28
Conclusions

• Successes
• Challenges
• Parting thoughts

29
Successes

• Research
• Attempts to model emotions are motivating
• Terminological clarifications
• Construction of structural, causal models
  (appraisal theories)
• Increasing interaction among experimentalists,
  modelers, and theorists
• Convergence on architecture structure
• Sloman, Ortony et al., Arbib-Fellous
• Principled analysis of representational /
  inferencing requirements
• Applications
• Many shallow models enhancing HCI and agents
• Sensing recognition expression

30
Challenges

• Data
• Emotion experiments are difficult
• esp. at the wetware level
• What can we learn from wetware
• Does neuromodulation translate into high-level
  parameters?
• Theories to guide model building
• Appraisal, regulation, mechanisms of emotion
  effects..
• Model development
• More principled strategies needed
• Can we build LTMs or do these have to evolve
  through interactions with their environment
  (Matthews, 2004)
• Validation
• Verification is not validation (Levis, 2003)
• Developing validation criteria

31
Parting Thoughts

• Neuroscience evidence indicates that emotion and
  cognition function as integrated systems
• There is no emotional homunculus (Fellous,
  2005)
• Emotions appear to perform useful and necessary
  functions in animals
• Understanding emotions helps us to identify these
  functions and their mechanisms
• Agents need these types of functions for
  effective, adaptive behavior
• BUT - does that mean agents need emotions?
• Goal management need not be emotional
• Does reward and punishment in agents require
  emotions?
• Are emotions specific to wetware or do they
  represent universal processes necessary for
  functioning in complex, uncertain environments?

32
Parting Thoughts

• Emotions as the ultimate integrating mechanism?
• Coordinate type 2 controls into a coherent type 1

33
The Rationality of Emotion and the
emotionality of reason
Eva Hudlicka Psychometrix Associates Blacksburg,
VA evahud_at_earthlink.net
MICS Symposium Saratoga Springs, NY March 4-6,
2005