Transcript This Time with Feeling: Modeling Effects of Emotion on Decision

From cognitive biases to panic:
Modeling the mechanisms of
anxiety disorders
Eva Hudlicka
Psychometrix Associates / U.Mass - Amherst
Amherst, VA
[email protected]
psychometrixassociates.com
Workshop on
“Computational Modeling of Cognition-Emotion Interactions”
CogSci 2014, Quebec City, Canada
Outline
• Affective biases on cognition  anxiety disorders
• Modeling Context:
– Cognitive-Affective Symbolic Architecture
– Search & rescue task
• Approach: Affective biases as architecture
parameters
• Example
• Implications for psychotherapy
2
Affective Biases
• Emotion effects on cognition can improve
…or degrade performance
• e.g., Anxiety-induced threat bias
– Adaptive:
– Maladaptive :
vigilance
anxiety & panic
3
Modeling Anxiety Effects:
The Good, the Bad & the Ugly
• Anxiety effects on cognition:
–
–
–
–
Attentional narrowing
Bias toward detection of threatening stimuli
Bias toward interpretation of ambiguous stimuli as threats
Promotion of self-focus
Adaptive
vigilance
Trait-anxious
over-protective
behavior
the Good
the Bad
Anxiety disorders &
panic attacks
same underlying mechanisms?
the Ugly
4
Benefits of Modeling
• Enable construction of alternative mechanisms for
observed effects
• Understand etiology of affective disorders
• Facilitate mechanism-based diagnosis
(beyond DSM-5 descriptions)
• More customized / targeted treatment
– Computer-based tools (serious games)
– Modeling the ‘patient’ ?
5
Context
• Symbolic cognitive-affective architecture
• Models high-level decision-making
• Models both emotion generation & emotion effects
• Emotion effects modeled in terms of parameters
controlling architecture processing
• Architecture controls agent behavior
… within a search & rescue team task
6
Task Context
- Search & rescue task in Arctic terrain
- Snowcat drivers (starting in lower left) trying to reach “Lost
Party” (red, upper right)
- Supply stations along routes
- Emergency tasks create obstacles & trigger stress
7
Task Context
Lost
Party
Supply
Station
Emergency
Task
Snow Cat
8
MAMID Architecture: Semantics & Data Flow
Cues: State of the world
(“Emergency task within range”
“Resources adequate”)
Cues
Attention
Situations: Perceived state
( “Able to process task” )
Situation
Assessment
Expectations: Expected state
Expectation
Generation
(“Task successfully completed”;
“Game points gained”; “Game won”)
Emotion
Generation
Affective state & emotions:
Happiness: High
Anxiety:
Low
Goals: Desired state
(“Game points = high”)
Goal
Manager
Action
Selection
Actions
Actions: to accomplish goals
(“Process Emergency Task”)
9
Modeling Emotion Effects via
Parameters Controlling Cognition
EMOTIONS /
TRAITS
ARCHITECTURE
PARAMETERS
Processing
Module Parameters
Traits
Extraversion
Neuroticism
Conscientiousness
Aggressiveness
speed, capacity
Attention
Situation
Assessment
Construct parameters
Cue selection & delay
….
Structural
Emotions
Anxiety
Anger
Sadness
Joy
COGNITIVE ARCHITECTURE
Architecture topology
Data flow among
modules
Long-term memory
Content & structure
Expectation
Generation
Emotion
Generation
Goal
Manager
Action
Selection
10
Modeling Threat Bias
TRAITS /
STATES
COGNITIVE
ARCHITECTURE
PARAMETERS
Processing
Parameters
Threat constructs
rated more highly
Traits
Neuroticism
Predisposes towards
Higher
Anxiety / Fear
Attention
Process
threat cues
Situation
Assessment
Module &
Construct parms.
- Cue selection
- Interpretive biases
Preferential processing of
...
Threatening stimuli
Emotions
COGNITIVE ARCHITECTURE
Process
threatening
Expectationinterpretations
Generator
Affect
Appraiser
Goal
Manager
Action
Selection
11
Modelling Panic Attack
• High state of anxiety induces a “perfect storm” of
biases
– Extreme threat bias
– Extreme self bias
– Reduced attention capacity
• Limited capacity precludes processing of useful cues
& derivation of alternative interpretations of situations
• No goals or actions generated
• Resulting behavioral paralysis further increases
anxiety
12
Internal Processing During a
Panic Attack
- Snowcat driver encounters an “Emergency Task”
while running low on supplies
13
Internal Processing During a
Panic Attack
Anxiety level is high
ANXIETY
High anxiety level
causes low
processing capacity
14
Internal Processing During a Panic
Attack: Mental Constructs in
Architecture Module Buffers
Attention: High
threat & emotion
cues only
SA: Negative
situations only
Goal Manager: No
goals selected
Behavior Selection: No
action selected due to
(a) extreme self focus;
(b) no goals
15
Modelling Alternative Mechanisms
of Anxiety & Panic Attacks
• Multiple, interacting causal pathways
… for each type of bias
• Parameter values are linear combinations of weighted factors
– (Wfactor1 * factor1) + (Wfactor2 * factor2) …
High Anxiety Intensity
Higher Sensitivity to
Anxiety
Lower baseline
attention capacity
Reduced attention
capacity
16
Alternative
Mechanisms
for
- Replace anxiety-generating belief net cluster with a cluster
from ‘Happy’ agent
Increasing
Attention
Capacity
- Change agent’s ‘beliefs’ – e.g., cognitive therapy
Modify emotion generation to derive lower anxiety intensity:
- Quantify contributions of specific beliefs
- Lower anxiety intensities--> Higher capacity values --> More Cues
Lower Anxiety Intensity
Lower Sensitivity to
Anxiety
Increase fundamental
attention capacity
Increased attention
capacity
17
Alternative Mechanisms for
Increasing Attention Capacity
Reduce sensitivity to anxiety via physiological manipulations
- Psychotropic medications
- Exercise
- Mindfulness
Lower sensitivity  Lower anxiety  Higher capacity  More cues
Lower Anxiety Intensity
Lower Sensitivity to
Anxiety
Increase fundamental
attention capacity
Increased attention
capacity
18
Implications for Psychotherapy
• Identify pathway(s) contributing to anxiety
– Specific (distorted?) beliefs?
– Increased baseline sensitivity?
• Target specific pathways.. via customized treatment
environments
– Virtual reality
– Serious games
• …possibly?… build model of patient within a particular
context (e.g., serious gaming)
• (Dis)confirm mechanism-based diagnosis via
modeling
19
Parting Thought
20