ROBOTS AND EMOTIONS
COMPUTATIONALLY MODELLING HUMAN EMOTION (MARSELLA)
EMOTIONS
Powerful due to shared knowledge of motivational power of emotions
Inherently dynamic & rapidly changing
Processes similar to emotion required by any intelligent entity facing social
environments
Functions of emotions
Interrupting & re-prioritising cognition
Improving multi-agent coordination & group utility – emotion & expression help
signal mental states
As motivator
COMPUTATIONAL MODELS
Research into computational models influences theories of human emotions
Processes & interactions of theories must be formally detailed to realise theory as
computational model
Implicit assumptions & hidden complexities are exposed
Conceptual level – computational models enrich language-of-emotion theories
Empirical level – computational models facilitate expanded range of predictions
Simulations way to explore temporal dynamics of emotion processes & form
predictions about source and time course of those dynamics
Challenges for computational models of emotion
Address how emotions arise & evolve over a range of conditions
Emotional responses can be rapid or unfold over minutes / days
FROM THEORY TO MODEL
Many different psychological theories of emotion – can be put into 3 categories
Discrete theories Limited number of core emotions that are biologically determined & innate
of emotion Their expression is shared across people & cultures
Dimensional Emotion & other affective phenomena should be conceptualised as a
theories of point in continuous space
emotion Discrete emotion categories do not have a specific biological basis
(brain region)
Computational models often use PAD theory – 3 dimensions
(pleasure, arousal, dominance)
Appraisal Emotions arise from process of comparing individual needs to
, theories of external demands
emotion Emotions reflect person-environment relationship
Cannot be explained only by environment OR only by
individual
Two people associate different emotions with the same external
stimulus
Dominant framework for building computational models of
emotions
Criticism
Emphasis on role of inferential processes (most attractive to
computational modelers & most controversial feature within
psychology)
Emotion is inherently reactive, appraisal are viewed as
consequent NOT precursor of emotional reactions
THE EMA (EMOTION & ADAPTATION) MODEL
Computational model of appraisal includes
Appraisal-derivation process – interpret representation of person-environment
relationship to derive a set of appraisal variables
Emotion-derivation model – takes this set of appraisal & produces an emotional
response
Behavioural consequence processes / coping strategies – triggered by emotion &
manipulate person-environment
PERSON-ENVIRONMENT RELATION
Serves as both input to & output of various appraisal processes
Agent’s view of how it relates to the environment
Causal interpretation – encodes input, intermediate results, output of inferential
processes
Inferential processes – mediate between agent’s goals & its physical and social
environment
Agent’s interpretation of agent-environment relationship
Snapshot of agent’s current knowledge concerning agent-environment relationship
Need for rapid appraisal – requirement on inferential processes that maintain causal
interpretation
APPRAISAL-DERIVATION PROCESS
Assume appraisal is fast, parallel, automatic
All significant features in causal interpretation are appraised separately,
simultaneously, automatically
Appraisal process associates data structure (appraisal frame) with each proposition
COMPUTATIONALLY MODELLING HUMAN EMOTION (MARSELLA)
EMOTIONS
Powerful due to shared knowledge of motivational power of emotions
Inherently dynamic & rapidly changing
Processes similar to emotion required by any intelligent entity facing social
environments
Functions of emotions
Interrupting & re-prioritising cognition
Improving multi-agent coordination & group utility – emotion & expression help
signal mental states
As motivator
COMPUTATIONAL MODELS
Research into computational models influences theories of human emotions
Processes & interactions of theories must be formally detailed to realise theory as
computational model
Implicit assumptions & hidden complexities are exposed
Conceptual level – computational models enrich language-of-emotion theories
Empirical level – computational models facilitate expanded range of predictions
Simulations way to explore temporal dynamics of emotion processes & form
predictions about source and time course of those dynamics
Challenges for computational models of emotion
Address how emotions arise & evolve over a range of conditions
Emotional responses can be rapid or unfold over minutes / days
FROM THEORY TO MODEL
Many different psychological theories of emotion – can be put into 3 categories
Discrete theories Limited number of core emotions that are biologically determined & innate
of emotion Their expression is shared across people & cultures
Dimensional Emotion & other affective phenomena should be conceptualised as a
theories of point in continuous space
emotion Discrete emotion categories do not have a specific biological basis
(brain region)
Computational models often use PAD theory – 3 dimensions
(pleasure, arousal, dominance)
Appraisal Emotions arise from process of comparing individual needs to
, theories of external demands
emotion Emotions reflect person-environment relationship
Cannot be explained only by environment OR only by
individual
Two people associate different emotions with the same external
stimulus
Dominant framework for building computational models of
emotions
Criticism
Emphasis on role of inferential processes (most attractive to
computational modelers & most controversial feature within
psychology)
Emotion is inherently reactive, appraisal are viewed as
consequent NOT precursor of emotional reactions
THE EMA (EMOTION & ADAPTATION) MODEL
Computational model of appraisal includes
Appraisal-derivation process – interpret representation of person-environment
relationship to derive a set of appraisal variables
Emotion-derivation model – takes this set of appraisal & produces an emotional
response
Behavioural consequence processes / coping strategies – triggered by emotion &
manipulate person-environment
PERSON-ENVIRONMENT RELATION
Serves as both input to & output of various appraisal processes
Agent’s view of how it relates to the environment
Causal interpretation – encodes input, intermediate results, output of inferential
processes
Inferential processes – mediate between agent’s goals & its physical and social
environment
Agent’s interpretation of agent-environment relationship
Snapshot of agent’s current knowledge concerning agent-environment relationship
Need for rapid appraisal – requirement on inferential processes that maintain causal
interpretation
APPRAISAL-DERIVATION PROCESS
Assume appraisal is fast, parallel, automatic
All significant features in causal interpretation are appraised separately,
simultaneously, automatically
Appraisal process associates data structure (appraisal frame) with each proposition
