Towards the Generation of Visual Qualia in Artificial Cognitive Architectures
description
Transcript of Towards the Generation of Visual Qualia in Artificial Cognitive Architectures
Towards the Generation of Visual Qualia in Artificial Cognitive Architectures
Raúl Arrabales, Agapito Ledezma, Araceli SanchisCarlos III University of Madrid
Computer Science Department
BRAIN INSPIRED COGNITIVE SYSTEMS14 – 16 July 2010, Madrid, Spain.
Third International ICSC Symposium onModels of Consciousness (MoC 2010)
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
2
www.Conscious-Robots.com
Main Objective
Explore the possibility of specification of the content of visual qualia using a
computational model based on the Global Workspace Theory.
3
4
The “redness” of red
5
How are qualia generated?
www.Conscious-Robots.com
Qualia in Humans
6
LIGHT
Retina
Spike Stream
Sensation
?
BRAINSENSES MIND
www.Conscious-Robots.com
Qualia in Humans
7
LIGHT
Retina
Spike Stream
Sensation
?
BRAINSENSES MIND
How are sensations produced?
www.Conscious-Robots.com
The Mind-Body Problem
8
MaterialObservable
ImmaterialPrivate
BRAIN MIND
www.Conscious-Robots.com
Dimensions of Consciousness
9
Phenomenal Consciousness
Functional Consciousness
Respo
nses Stimuli
www.Conscious-Robots.com
Dimensions of Consciousness
10
Phenomenal Consciousness
Functional Consciousness
Respo
nses Stimuli
Qualia
“Hard Problem”
“Easy Problems”
www.Conscious-Robots.com
Dimensions of Consciousness
11
Phenomenal Consciousness
Functional Consciousness
Respo
nses Stimuli
Qualia
“Hard Problem”
“Easy Problems”
www.Conscious-Robots.com
What are qualia?
12
Private
IneffableIntegrated
Structured
Qualia
Presence“The redness of red”
“Hard Problem”
Sensations
“A headache”
“The flavor of an ice-cream”
“Enjoying a song”
www.Conscious-Robots.com
Why are qualia so elusive?
13
A B“Red” “Red”
www.Conscious-Robots.com
How to study phenomenology?
Heterophenomenology (Dennett, 1991).
14
1st Person Observations
Qualia
www.Conscious-Robots.com
How to study phenomenology?
Heterophenomenology (Dennett, 1991).
15
3rd Person Observations
2nd Person Observations
1st Person Observations
Inspection Report
Qualia
www.Conscious-Robots.com
How to study phenomenology?
Heterophenomenology (Dennett, 1991).
16
3rd Person Observations
2nd Person Observations
1st Person Observations
Inspection Report
Qualia
www.Conscious-Robots.com
Machine Consciousness
17
Human Consciousness Models
Machine Consciousness Models
Human Consciousness
Machine Consciousness
Analysis and Modeling
Design and Implementation
Comparison (Synthetic Phenomenology)
Adaptation to Computational
Models
PhysicalNeurophysiologicCognitive
Artificial Neural NetworksHybrid SystemsCognitive Architectures
www.Conscious-Robots.com
How to study phenomenology?
18
2nd Person Observations
1st Person Observations
Report
Qualia
2nd Person Observations
1st Person Observations
Report
Qualia
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
19
www.Conscious-Robots.com
Working Hypotheses about Qualia
They are related to cognitive functions.
Their contents have a functional role.
They are the ultimate outcome of the perception process.
20
www.Conscious-Robots.com
Proposed Model
21
Perceptual Content
Proprioceptive Sensing
Exteroceptive Sensing
Stage 1Perceptual
Content Representation
Stage 2Introspective
Perceptual Representation
Stage 3Self-Modulation and
Report
Sensory Data Visual Sensors(dot stimulus)
Somatosensory System
(sensor positions)
World Reconstruction
Introspection
Modulation / Reportability
Meta-Representation
Meta-Management
www.Conscious-Robots.com
Application to Visual Experience
22
150 ms 150 ms10 ms 10 ms
www.Conscious-Robots.com
Proposed Model
23
Perceptual Content
Proprioceptive Sensing
Exteroceptive Sensing
Stage 1Moving dot
Stage 2What is it like to see a moving dot
Stage 3I report to be watching
a moving dot
Sensory Data(left dot – blank – right
dot – blank)
Visual Sensors(dot stimulus)
Somatosensory System
(sensor positions)
World Reconstruction
Introspection
Modulation / Reportability
Meta-Representation
Meta-Management
www.Conscious-Robots.com
GWT Computational ModelGlobal Workspace Theory (Baars, 1988, 1997).
24
WorkingMemory(Scene)
SpecializedProcessors(Audience)
Spotlight
Context Formation and Executive Guidance(Director, scene designer, etc. behind the scenes)
Interim coalition
BroadcastBroadcast
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
25
www.Conscious-Robots.com
CERA-CRANIUM
A framework for experimentation with cognitive models of consciousness.
26
CERA-CRANIUM
Agent
Model
Sensors Actuators
www.Conscious-Robots.com
CERA-CRANIUM
CERA (Conscious and Emotional Reasoning Architecture)
Layered Control Architecture
CRANIUM (Cognitive Robotics Architecture Neurologically Inspired Underlying Manager)
Runtime Environment for the creation and management of specialized processors sharing a global working memory.
27
www.Conscious-Robots.com
CERA-CRANIUMMinimal Implementation
What should be the next action of the agent?What should be the next “conscious” content of the agent?
28
Physical LayerMission-specific
LayerCore Layer
SensorServices
MotorServices
Sensors
Actuators
CERAROBOT
…
CRANIUM Workspace
Single Percepts
Complex Percepts
Sensor Service
Commands
Core L
ayer
CERA Viewer
www.Conscious-Robots.com
CERA-CRANIUM Observer
29
CERA. Core Layer
(focus onsaliencies)
CERA. Physical Layer
Sensor Service
Sensor Service
Simple Percepts
…
CRANIUM Workspace
Complex Percepts
Pre-processors
…
Sensor Service
CERA. S-MSensors
…
Aggregators
www.Conscious-Robots.com
CERA-CRANIUM Observer
30
CERA. Core Layer
(focus onsaliencies)
CERA. Physical Layer
Sensor Service
Sensor Service
Simple Percepts
…
CRANIUM Workspace
Complex Percepts
Pre-processors
…
Sensor Service
CERA. S-MSensors
…
Aggregators
Percept Aggregators
Complex Percepts
M(SCJ)
Simple Percepts
Sensor Preprocessors
Timer
Proprioception
N(δSj)N(δSJ)
j t
Sensor Readings
www.Conscious-Robots.com
CERA-CRANIUM Observer
31
Working Memory(GW)
Raw Monomodal Sensory Data
“Spotlight”
Context Formation ProcessesCoordination Processes
Sensors
Control Signal
Asynchronous InputHigh Bandwidth
Sequential OutputLow Bandwidth
Context ArtificialQualia
Integrated Multimodal Representations
Specialized Processors
GOALS
www.Conscious-Robots.com
Contextualization
Bottom-Up: Native Spatiotemporal contexts.
Top Down: Specific contexts induced from the Core
Layer.
32
www.Conscious-Robots.com
Contextualization
33
www.Conscious-Robots.com
Contextualization
34
Single Percepts
Complex Percept
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
35
www.Conscious-Robots.com
Specialized Processors
Region of Interest detector for white objects.
Motion Detector.
36
www.Conscious-Robots.com
Visual Experience
37
CERACRANIUM
Visual Stimuli Human
Observer
Robot Cam
“I see an object moving downwards”
CERA Viewer
Artificial Qualia
Specification
Content Specification can
be directly compared
www.Conscious-Robots.com
Visual Experience
38
Visual Stimuli: S1: Static white object in a dark background. S2: White object moving along a rectilinear trajectory. S3: Two stationary white blinking rounded spots.
CERACRANIUM
Visual Stimuli Human
Observer
Robot Cam
“I see an object moving downwards”
CERA Viewer
Artificial Qualia
Specification
Content Specification can
be directly compared
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
39
www.Conscious-Robots.com
Preliminary Results
40
(a)
“Objet resting on the ground”
(b)
“Object moving uniformly from the right to the
left”
“Ball moving back and forth from the left to
the right”
(c)
RDS SIMULATOR
SIMULATED CAM
CERA VIEWER
S1
S2
S3
www.Conscious-Robots.com
Contents Introduction Computational Model CERA-CRANIUM Experimental Setting Preliminary Results Conclusions Future Work
41
www.Conscious-Robots.com
Conclusions
Using GWT will shed light on whether or not the model can account for typical human perceptual effects.
Synthetic Phenomenology might help us understand qualia.
For instance:
Does the presence of perceptual illusions correlates with better perception accuracy in noisy environments?
42
www.Conscious-Robots.com
Future Work
More complex stimuli. Multimodal stimuli. Real world scenarios.
Better specification and representation of the content of Artificial Qualia.
Improve the Cognitive Architecture: Expectations. Emotions. …
43
www.Conscious-Robots.com
Thank you for your attention. Any questions?
44