Possible Solutions from the Cognitive Neuroscience of Emotion
David Sander Geneva Emotion Research Group University of Geneva
A role for CN in designing emotion-oriented systems?
Levels of analyses in CN Problems, and CN directions Artificial emotions Recognition of facial expression
What is CN?
The emergence of a discipline
Cognitive Neuroscience Institute (Dartmouth): 1982 Journal of Cognitive Neuroscience: 1988 Cognitive Neuroscience Society: 1993 Institute of Cognitive Neuroscience (London): 1996 the
task of cognitive neuroscience is to map
the information-processing structure of the human mind and to discover how this computational organization is implemented in the physical organization of the brain Tooby & Cosmides (2000)
Levels of analyses in CN The perils of sitting on a one-legged stool
(Kosslyn & Intrilligator, 1992)
Only one paradigmatic leg a stability peril for the model
Informationprocessing model
B E H A V I O R
Many Psychological models are sitting only on a behavioral account
Levels of analyses in CN The perils of sitting on a one-legged stool
(Kosslyn & Intrilligator, 1992)
Only one paradigmatic leg a stability peril for the model
Informationprocessing model
B R A I N
Many Neurobiological and some neuropsychological models are sitting only on a brain account
Levels of analyses in CN The perils of sitting on a one-legged stool
(Kosslyn & Intrilligator, 1992)
Only one paradigmatic leg a stability peril for the model
Informationprocessing model
C O M P U T A T I O N
Many Artificial Intelligence models are sitting only on a computational account
The advantage of sitting on a three-legged stool
Three paradigmatic legs more stability for the model
Informationprocessing model
C O M P U T A T I O N
B E H A V I O R
B R A I N
Ideal CN models are sitting on behavioral, brain, and computational accounts
Cognitive Neuroscience Triangle
Behavior
Computation
Analyses Models
Brain
Neural Activity (Neurophysiology) Areas & Connections (Neuroanatomy)
Problems, and CN directions: Problem 1 Emotion-oriented system, but... ...oriented towards which level?
Behavioral
An artificial behaviorally believable output response given a natural input, whatever the plausibility of the architecture
Computational (or representational) Neural
Other (?)
Problem 1
Natural Processes versus Artificilal Efficiency
Is it important to know how the human brain computes emotion in order to develop a humaine emotion-oriented system?
Appraisal of a threat, Autonomic activity, Withdrawing, Expression, and Feeling of being afraid
Behavioral plausible output: humaine Autonomic activity, emotion-oriented Withdrawing, system Expression of fear.
Problem 1
Emotion-oriented system, but...
...oriented towards which level?
Behavioral
An artificial behaviorally believable output response given a natural input, whatever the plausibility of the architecture
Computational (or representational)
An artificial system that is constrained by the functional architecture designed by CN results
CN is useless
Problem 2
Selecting the functional architecture to be implemented in an artificial emotion system i. Dissociation of emotional processes ii. Implementation of emotional processes in the brain iii. Time course of emotional processes
Problem 2: selecting the functional architecture
Three main approaches: Basic Emotions Approach Dimensional Approach Systems-level Approach
Problem 2: selecting the functional architecture
CN and Basic Emotions
Most of the past Cognitive Neuroscience researches on emotion focused on the attempt to find specific brain regions implementing discrete basic emotions: The various classes of emotion are mediated by separate neural systems (...) (LeDoux, 1996)
Problem 2: selecting the functional architecture
CN and Basic Emotions
hman & Mineka (2001):
The amygdala is a fear module
Basica!y, the fear module is a device for activating defensive behaviour and associated psychophysiological responses and emotional feelings to threatening stimuli.
Panksepp (2003)
237-239
Problem 2: selecting the functional architecture
CN and the Dimensional Approach
Some recent Cognitive Neuroscience researches were interested in dissociating the dimensions of Valence and Intensity (Anderson et al., 2003; Small et al., 2003). (!! Intensity Activation !!)
Valence versus Intensity
Anderson et al. (2003), Nature Neuroscience
Problem 2: selecting the functional architecture
CN at the systems-level
Some CN researchers take into consideration the complexity of emotion by parsing its subcomponents at the systems-level and, sometimes, by attempting to model the interactions between the proposed processes: Action tendencies (e.g., Davidson) Somatic signals (e.g., Damasio) Feeling (e.g., Lane)
Action tendencies (e.g., Davidson, 1995)
Perception/Production
distinction between perception of the emotional value of a stimulus versus the production of expressive behavior
Anterior activation asymmetry model Left anterior region associated with approachrelated emotions Right anterior region associated with withdrawalrelated emotions
Somatic signals (e.g., Damasio, 1998)
A critical function of somaticrelated signals and their integration with the other brain signals.
Feeling
Feeling as an integration of some emotional signals
The conscious experience is integrated via a convergence zone that could be the Anterior Cingulate and/or the Medial Prefrontal Cortex (Reiman. 1997; Lane, 2000). The subjective feeling is integrated via the synchronization of other components (Scherer, 2003). Binding through synchronization was proposed for the visual system for example.
Appraisal Theory
Relevant (e.g., unpleasant, goal obstructive), Dicult to cope with Event
Appraisal Processes Emotional Expression
Autonomic activation
Action Tendencies Subjective Feeling
Withdrawal I am a$aid
Cognitive Neuroscience of Appraisal Processes Event
High level exteroceptive processing Implication
Integrative cortices
Normative Significance Goal representation
Con d e p text end enc e
Sensory cortices
DLPFC Hippo OFC
Intrinsic pleasantness
Action tendency
MPFC ACC
Regulation, coping
Sensory Thalamus
Coarse exteroceptive processing
Ventral Striatum
Motivational bases (reward)
Relevance detection
Amy
Somatosensoryrelated cortical Somatic maps and subcortical structures
Emotional expression
Sander & Scherer, in prep.
Neuroendocrine/Autonomic/Somatic NS Body state
Problem 3
Recognition of facial expression
(from Haxby et al., 2000)
Colliculus-pulvinar-amygdala Pathway
LGB: Lateral Geniculate Body SC: Superior Colliculus V1: Primary Visual Cortex Pulvinar
Visual Cortex
V1
Amygdala
SC
LGB
Retina
Recognition of a facial expression of fear
Stimulus 120 ms: Fast early processing of highly relevant events
A, amygdala; FFA, fusiform face area; INS, insula; O, orbitofrontalcortex; SC, superior colliculus; SCx, striate cortex; SS, somatosensorycortex; STG, superior temporal gyrus; T, thalamus.
From Adolphs (2002). Current Opinion in Neurobiology
170 ms: - detailed perception; - emotional reaction involving the body
> 300ms: Conceptual knowledge of the emotion signaled by the face
Emotion-oriented system, but...
...oriented towards which level?
Behavioral
An artificial behaviorally believable output response given a natural input, whatever the plausibility of the architecture
Computational (or representational) Neural
An artificial system that is constrained by the An artificial system that functional architecture is constrained by the designed by CN results functional architecture and natural neural networks properties
CN is useless
CN can help
CN can help
Problem 4
Multimodal integration
Timing: Results suggest that audio-visual emotional binding is early in time (110 ms post-stimulus) Integrative structure
Amygdala response to congruent fearful voices and faces Dolan et al. (2001)
->Test of multimodal emotion display in ECA using brain-imaging
Problem 5
Influence of dynamism in the facial expression on perceived emotion
Emotion morphs depicted expression changes of getting scared or getting angry in real-time. Brain regions implicated in processing facial affect, including the amygdala and fusiform gyrus, showed greater responses to dynamic versus static emotional expressions.
Labar et al. (2003)
Conclusion
Cognitive Neuroscience can help to find solutions for emotionoriented systems mainly if they are focused on the computational, and/or the neural levels. Artificial emotions: A decisive choice between: as many systems as emotions different systems for approach-related versus withdrawalrelated emotions a system for intensity, a system for valence (but, only feeling) a system for each emotional component Recognition of emotional expression: Modeling two pathways (one for coarse and fast processing, and one for detailed proc.). A computational model of emotional processes would benefit from modeling other closely related cognitive processes, such as attention.
