SlideShare a Scribd company logo
Creating Consciousness
Ryota Kanai
(Araya, Inc.)
Araya, Inc.
Our Mission
We want to understand consciousness by creating
it.
Our Research
• Integrated Information Theory (Oizumi, Kitazono)
• Intrinsic Motivation (Biehl, Magrans)
• Free Energy/ VAEs (Yu, Tamai)
• Coarse Graining (Chang)
• Meta-learning (Guttenberg)
• Wasserstein distance (Amari, Oizumi, Mizutani)
Research Themes
• Integrated Information
Theory (Oizumi, Kitazono)
• Intrinsic Motivation (Biehl,
Magrans)
• Free Energy/ VAEs (Yu,
Tamai)
• Coarse Graining (Chang)
• Meta-learning (Guttenberg)
• Wasserstein distance
(Amari, Oizumi, Mizutani)
We want to understand consciousness by creating
it.
Our Mission
YHouse:
yhousenyc.org
AI and Society:
aiandsociety.org
YHouse:
conresnet.org
• Consciousness is relevant for all kinds of cognition.
– Learning and Memory, Perception, Thoughts, Action, Decision
Making, Emotion…
• We don’t understand understanding
– What does it mean to understand?
– What does it mean to feel something?
– What is consciousness for?
Why should we care about consciousness?
1. Spontaneous behavior (e.g. Intention, Curiosity)
2. Generalization (e.g. One-shot learning, Creativity, Thought)
3. Explainability (e.g., Metacognition, Language)
Current issues in AI
Current challenges in AI
A huge gap from current AI to AGI
We need to understand consciousness for AGI
Four kinds of machine
consciousness
• MC1. Machines with the external behaviour associated
with consciousness.
• MC2. Machines with the cognitive characteristics
associated with consciousness.
• MC3. Machines with an architecture that is claimed to be a
cause or correlate of human consciousness.
• MC4. Phenomenally conscious machines.
Gamez (2008)
• How can we prove that it has phenomenal
experience?
Proving Consciousness
The dual problem of machine consciousness
Creating Consciousness
• How can we create a machine that reproduces
causal structures and functions of consciousness?
Functions of consciousness
• Depiction. The system has perceptual states that ‘represent’ elements of
the world and their location.
• Imagination. The system can recall parts of the world or create
sensations that are like parts of the world.
• Attention. The system is capable of selecting which parts of the world to
depict or imagine.
• Planning. The system has control over sequences of states to plan
actions.
• Emotion. The system has affective states that evaluate planned actions
and determine the ensuing action.
E.g. Aleksander (2005)
Example: Aleksander’s Axioms
Consciousness for bridging a temporal gap
Clark & Squire (1998)
• Delay conditioning occurs without awareness
• Trace conditioning requires awareness (and hippocampus)
Online and memory guided action
Perceptual judgement
Visually guided action
Milner et al. (1991)
Memory guided action depends on perceptual system
V1 hypothesis
• V1 is not conscious because it does not project to
prefrontal cortex.
• The rationale from biological usefulness of awareness
– A: To produce the best current interpretation of the visual scene
– B: To make the information available for the system that plan
and execute voluntary motor outputs.
Crick & Koch (1995)
A New Hypothesis on
Consciousness
A Counterfactual Information Generation Hypothesis
Agents with the ability to generate predictions, including
counterfactual future and past, of their own state in the
environment have consciousness.
Implementing
counterfactuals
Active inference
Friston et al. (2010) Biol Cybern
The motivation: Biological agents need to minimise the entropy of their
sensory states to resist a tendency to disorder.
(Ergodic Assumption)
(Entropy of sensory states)
This implies that agents should minimise surprise over time.
(Average over space = Average over time)
Active inference
c.f. series of work by Friston
Generative model
Active inference
Perceptual Inference
Active inference with counterfactuals
c.f. series of work by Friston
Instead of just minimising the entropy of sensory states, more recent formulations
suggest minimising the joint entropy of sensory and hidden states.
We can decompose this into:
(Entropy of Sensory States) + (Conditional Entropy of Hidden States)
Counterfactual predictions
Next eye positions are chosen to
minimise uncertainty of hidden
states
Counterfactual predictions in a robot
Bongard , Zykov & Lipson (2006)
• Intention
– Counterfactual predictions of the consequences of future actions.
• Non-Reflexive Behaviour
– Mental simulation detached from the present external stimuli.
• Perceptual Presence (Seth 2014)
– Coherent sensory motor contingency.
Summary
A model of the world, and sensory motor contingency are
prerequisites for the ability to simulate the world internally.
Function of Consciousness = The ability to represent
events disconnected from the present environment.
What does it mean to generate information?
input
encoding decoding
output
VAE: µ x σ  z
Info Processing Info Generation
For individual layers, the analogy represented here is not obvious.
Hidden: z
Link to predictive coding
Prediction error
Info. Processing
Prediction
Info. Generation
input
Hidden: z
ConsciousUnconscious
DecodingEncoding
GenerationProcessing
DecompressionCompression
PredictionPrediction Error
FeedbackFeedforward
Feedback generates information, hence consciousness.
Feedback and visual awareness
Super et al. (2001)
Super et al. (2001)
Late activity (feedback) is associated with subjective visibility.
Feedback and visual awareness
Optogenetics evidence
Manita et al. (2015) Neuron
Creating an agent with
information generation capability
Creating generative agents
Generative control:
 'What is the distribution of possible futures?'
Distribution includes actions → search for actions that lead to desirable
futures
 Can change reward function on the fly
 Reward function can easily depend on things like uncertainties and
surprisals
 Free long-term planning/coherency
Latent
Space
Future
Sensory
State
Future
Action
Pattern
Latent
Space
Latent
Space
Latent
Space
Latent
Variable
Future
Sensory
State
Future
Action
Pattern
Future
Sensory
State
Future
Action
Pattern
Future
Sensory
State
Future
Action
Pattern
Future
Sensory
State
Future
Action
Pattern
Generative Model
Generative control
Repeated re-encoding allows an auto-encoder to better match the data distribution
(Arulkumaran, Creswell, Bharath 2016)
Recurrent Autoencoder
Generative Control of Cartpole
Learn action+sensor generative model, maps latent variable → prediction
for next 16 frames. Gradient descent in latent space to pick action
Ultimate method: Direct connection
• Qualia seem to be shared among the brains through direct
connections.
• We can connect brains directly to AI to check the qualia in
AI.
The Hogan sisters seem to share qualia
via direct connections in the thalamus.
Final thoughts
• Information generation, not processing
– We should pay more attention to the generative aspect of the brain.
• Qualia are not counterfactual
– Perception of the present is also generated.
– Corresponding bottom-up signals may help generation of more detailed, clear
contents.
• Intrinsic definition of information generation
– IIT-like definition?
• Phenomenal consciousness
– If we create the causal structure of the third-person aspect of consciousness,
the first-person aspect of consciousness should be generated (Biological
Naturalism).
Thank you
Ryota Kanai
kanair@araya.org
@kanair

More Related Content

Similar to Creating Consciousness

Embodied cognition
Embodied cognitionEmbodied cognition
Embodied cognition
Anna Sharkova MSc
 
Neural fields, a cognitive approach
Neural fields, a cognitive approachNeural fields, a cognitive approach
Neural fields, a cognitive approach
Nicolas Rougier
 
Artificial intelligent Lec 1-ai-introduction-
Artificial intelligent Lec 1-ai-introduction-Artificial intelligent Lec 1-ai-introduction-
Artificial intelligent Lec 1-ai-introduction-
Taymoor Nazmy
 
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
IOSR Journals
 
Psychology emotional design and IA
Psychology emotional design and IAPsychology emotional design and IA
Psychology emotional design and IA
Brian Cugelman, PhD (AlterSpark)
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Huahai Yang
 
Deep Visual Understanding from Deep Learning by Prof. Jitendra Malik
Deep Visual Understanding from Deep Learning by Prof. Jitendra MalikDeep Visual Understanding from Deep Learning by Prof. Jitendra Malik
Deep Visual Understanding from Deep Learning by Prof. Jitendra Malik
The Hive
 
alexVAE_New.pdf
alexVAE_New.pdfalexVAE_New.pdf
alexVAE_New.pdf
sourabhgothe1
 
Ch 1 Introduction to AI.pdf
Ch 1 Introduction to AI.pdfCh 1 Introduction to AI.pdf
Ch 1 Introduction to AI.pdf
KrishnaMadala1
 
Ai history to-m-learning
Ai history to-m-learningAi history to-m-learning
Ai history to-m-learning
Kyung Eun Park
 
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
Dr.ganesh Narasimhan
 
Artificial Intelligence
Artificial IntelligenceArtificial Intelligence
Sensorimotor Network Development During Early Postnatal Life in the Awake and...
Sensorimotor Network Development During Early Postnatal Life in the Awake and...Sensorimotor Network Development During Early Postnatal Life in the Awake and...
Sensorimotor Network Development During Early Postnatal Life in the Awake and...
InsideScientific
 
1 Introduction to AI.pptx
1 Introduction to AI.pptx1 Introduction to AI.pptx
1 Introduction to AI.pptx
BikashAcharya13
 
AI: Introduction to artificial intelligence
AI: Introduction to artificial intelligenceAI: Introduction to artificial intelligence
AI: Introduction to artificial intelligence
DataminingTools Inc
 
AI: Introduction to artificial intelligence
AI: Introduction to artificial intelligenceAI: Introduction to artificial intelligence
AI: Introduction to artificial intelligence
Datamining Tools
 
Neuroscienze e Libero Arbitrio
Neuroscienze e Libero ArbitrioNeuroscienze e Libero Arbitrio
Neuroscienze e Libero Arbitrio
Fondazione Giannino Bassetti
 
Artificial intelligence introduction
Artificial intelligence  introduction Artificial intelligence  introduction
Artificial intelligence introduction
San1705
 

Similar to Creating Consciousness (20)

Embodied cognition
Embodied cognitionEmbodied cognition
Embodied cognition
 
Neural fields, a cognitive approach
Neural fields, a cognitive approachNeural fields, a cognitive approach
Neural fields, a cognitive approach
 
Artificial intelligent Lec 1-ai-introduction-
Artificial intelligent Lec 1-ai-introduction-Artificial intelligent Lec 1-ai-introduction-
Artificial intelligent Lec 1-ai-introduction-
 
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
Universal Artificial Intelligence for Intelligent Agents: An Approach to Supe...
 
Are robots present?
Are robots present?Are robots present?
Are robots present?
 
Psychology emotional design and IA
Psychology emotional design and IAPsychology emotional design and IA
Psychology emotional design and IA
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's Perspective
 
Deep Visual Understanding from Deep Learning by Prof. Jitendra Malik
Deep Visual Understanding from Deep Learning by Prof. Jitendra MalikDeep Visual Understanding from Deep Learning by Prof. Jitendra Malik
Deep Visual Understanding from Deep Learning by Prof. Jitendra Malik
 
alexVAE_New.pdf
alexVAE_New.pdfalexVAE_New.pdf
alexVAE_New.pdf
 
Ch 1 Introduction to AI.pdf
Ch 1 Introduction to AI.pdfCh 1 Introduction to AI.pdf
Ch 1 Introduction to AI.pdf
 
Ai history to-m-learning
Ai history to-m-learningAi history to-m-learning
Ai history to-m-learning
 
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
AI Introduction Artificial intelligence introduction fundamentals alogirthms ...
 
Artificial Intelligence
Artificial IntelligenceArtificial Intelligence
Artificial Intelligence
 
Sensorimotor Network Development During Early Postnatal Life in the Awake and...
Sensorimotor Network Development During Early Postnatal Life in the Awake and...Sensorimotor Network Development During Early Postnatal Life in the Awake and...
Sensorimotor Network Development During Early Postnatal Life in the Awake and...
 
1 Introduction to AI.pptx
1 Introduction to AI.pptx1 Introduction to AI.pptx
1 Introduction to AI.pptx
 
AI: Introduction to artificial intelligence
AI: Introduction to artificial intelligenceAI: Introduction to artificial intelligence
AI: Introduction to artificial intelligence
 
AI: Introduction to artificial intelligence
AI: Introduction to artificial intelligenceAI: Introduction to artificial intelligence
AI: Introduction to artificial intelligence
 
Neuroscienze e Libero Arbitrio
Neuroscienze e Libero ArbitrioNeuroscienze e Libero Arbitrio
Neuroscienze e Libero Arbitrio
 
Artificial intelligence introduction
Artificial intelligence  introduction Artificial intelligence  introduction
Artificial intelligence introduction
 
Ai
Ai Ai
Ai
 

Recently uploaded

Free Complete Python - A step towards Data Science
Free Complete Python - A step towards Data ScienceFree Complete Python - A step towards Data Science
Free Complete Python - A step towards Data Science
RinaMondal9
 
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
Neo4j
 
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdfObservability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
Paige Cruz
 
Climate Impact of Software Testing at Nordic Testing Days
Climate Impact of Software Testing at Nordic Testing DaysClimate Impact of Software Testing at Nordic Testing Days
Climate Impact of Software Testing at Nordic Testing Days
Kari Kakkonen
 
UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5
DianaGray10
 
Elizabeth Buie - Older adults: Are we really designing for our future selves?
Elizabeth Buie - Older adults: Are we really designing for our future selves?Elizabeth Buie - Older adults: Are we really designing for our future selves?
Elizabeth Buie - Older adults: Are we really designing for our future selves?
Nexer Digital
 
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdfFIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
FIDO Alliance
 
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
James Anderson
 
Pushing the limits of ePRTC: 100ns holdover for 100 days
Pushing the limits of ePRTC: 100ns holdover for 100 daysPushing the limits of ePRTC: 100ns holdover for 100 days
Pushing the limits of ePRTC: 100ns holdover for 100 days
Adtran
 
Elevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object CalisthenicsElevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object Calisthenics
Dorra BARTAGUIZ
 
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
Neo4j
 
Introduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - CybersecurityIntroduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - Cybersecurity
mikeeftimakis1
 
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
James Anderson
 
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
Neo4j
 
Monitoring Java Application Security with JDK Tools and JFR Events
Monitoring Java Application Security with JDK Tools and JFR EventsMonitoring Java Application Security with JDK Tools and JFR Events
Monitoring Java Application Security with JDK Tools and JFR Events
Ana-Maria Mihalceanu
 
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdfFIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
FIDO Alliance
 
Securing your Kubernetes cluster_ a step-by-step guide to success !
Securing your Kubernetes cluster_ a step-by-step guide to success !Securing your Kubernetes cluster_ a step-by-step guide to success !
Securing your Kubernetes cluster_ a step-by-step guide to success !
KatiaHIMEUR1
 
The Future of Platform Engineering
The Future of Platform EngineeringThe Future of Platform Engineering
The Future of Platform Engineering
Jemma Hussein Allen
 
Generative AI Deep Dive: Advancing from Proof of Concept to Production
Generative AI Deep Dive: Advancing from Proof of Concept to ProductionGenerative AI Deep Dive: Advancing from Proof of Concept to Production
Generative AI Deep Dive: Advancing from Proof of Concept to Production
Aggregage
 
Microsoft - Power Platform_G.Aspiotis.pdf
Microsoft - Power Platform_G.Aspiotis.pdfMicrosoft - Power Platform_G.Aspiotis.pdf
Microsoft - Power Platform_G.Aspiotis.pdf
Uni Systems S.M.S.A.
 

Recently uploaded (20)

Free Complete Python - A step towards Data Science
Free Complete Python - A step towards Data ScienceFree Complete Python - A step towards Data Science
Free Complete Python - A step towards Data Science
 
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024GraphSummit Singapore | The Art of the  Possible with Graph - Q2 2024
GraphSummit Singapore | The Art of the Possible with Graph - Q2 2024
 
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdfObservability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
Observability Concepts EVERY Developer Should Know -- DeveloperWeek Europe.pdf
 
Climate Impact of Software Testing at Nordic Testing Days
Climate Impact of Software Testing at Nordic Testing DaysClimate Impact of Software Testing at Nordic Testing Days
Climate Impact of Software Testing at Nordic Testing Days
 
UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5UiPath Test Automation using UiPath Test Suite series, part 5
UiPath Test Automation using UiPath Test Suite series, part 5
 
Elizabeth Buie - Older adults: Are we really designing for our future selves?
Elizabeth Buie - Older adults: Are we really designing for our future selves?Elizabeth Buie - Older adults: Are we really designing for our future selves?
Elizabeth Buie - Older adults: Are we really designing for our future selves?
 
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdfFIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
FIDO Alliance Osaka Seminar: The WebAuthn API and Discoverable Credentials.pdf
 
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...
 
Pushing the limits of ePRTC: 100ns holdover for 100 days
Pushing the limits of ePRTC: 100ns holdover for 100 daysPushing the limits of ePRTC: 100ns holdover for 100 days
Pushing the limits of ePRTC: 100ns holdover for 100 days
 
Elevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object CalisthenicsElevating Tactical DDD Patterns Through Object Calisthenics
Elevating Tactical DDD Patterns Through Object Calisthenics
 
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
GraphSummit Singapore | The Future of Agility: Supercharging Digital Transfor...
 
Introduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - CybersecurityIntroduction to CHERI technology - Cybersecurity
Introduction to CHERI technology - Cybersecurity
 
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
Alt. GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using ...
 
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
GraphSummit Singapore | Graphing Success: Revolutionising Organisational Stru...
 
Monitoring Java Application Security with JDK Tools and JFR Events
Monitoring Java Application Security with JDK Tools and JFR EventsMonitoring Java Application Security with JDK Tools and JFR Events
Monitoring Java Application Security with JDK Tools and JFR Events
 
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdfFIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
FIDO Alliance Osaka Seminar: Passkeys at Amazon.pdf
 
Securing your Kubernetes cluster_ a step-by-step guide to success !
Securing your Kubernetes cluster_ a step-by-step guide to success !Securing your Kubernetes cluster_ a step-by-step guide to success !
Securing your Kubernetes cluster_ a step-by-step guide to success !
 
The Future of Platform Engineering
The Future of Platform EngineeringThe Future of Platform Engineering
The Future of Platform Engineering
 
Generative AI Deep Dive: Advancing from Proof of Concept to Production
Generative AI Deep Dive: Advancing from Proof of Concept to ProductionGenerative AI Deep Dive: Advancing from Proof of Concept to Production
Generative AI Deep Dive: Advancing from Proof of Concept to Production
 
Microsoft - Power Platform_G.Aspiotis.pdf
Microsoft - Power Platform_G.Aspiotis.pdfMicrosoft - Power Platform_G.Aspiotis.pdf
Microsoft - Power Platform_G.Aspiotis.pdf
 

Creating Consciousness

  • 2. Araya, Inc. Our Mission We want to understand consciousness by creating it. Our Research • Integrated Information Theory (Oizumi, Kitazono) • Intrinsic Motivation (Biehl, Magrans) • Free Energy/ VAEs (Yu, Tamai) • Coarse Graining (Chang) • Meta-learning (Guttenberg) • Wasserstein distance (Amari, Oizumi, Mizutani)
  • 3. Research Themes • Integrated Information Theory (Oizumi, Kitazono) • Intrinsic Motivation (Biehl, Magrans) • Free Energy/ VAEs (Yu, Tamai) • Coarse Graining (Chang) • Meta-learning (Guttenberg) • Wasserstein distance (Amari, Oizumi, Mizutani) We want to understand consciousness by creating it. Our Mission
  • 7. • Consciousness is relevant for all kinds of cognition. – Learning and Memory, Perception, Thoughts, Action, Decision Making, Emotion… • We don’t understand understanding – What does it mean to understand? – What does it mean to feel something? – What is consciousness for? Why should we care about consciousness?
  • 8. 1. Spontaneous behavior (e.g. Intention, Curiosity) 2. Generalization (e.g. One-shot learning, Creativity, Thought) 3. Explainability (e.g., Metacognition, Language) Current issues in AI Current challenges in AI A huge gap from current AI to AGI We need to understand consciousness for AGI
  • 9. Four kinds of machine consciousness • MC1. Machines with the external behaviour associated with consciousness. • MC2. Machines with the cognitive characteristics associated with consciousness. • MC3. Machines with an architecture that is claimed to be a cause or correlate of human consciousness. • MC4. Phenomenally conscious machines. Gamez (2008)
  • 10. • How can we prove that it has phenomenal experience? Proving Consciousness The dual problem of machine consciousness Creating Consciousness • How can we create a machine that reproduces causal structures and functions of consciousness?
  • 11. Functions of consciousness • Depiction. The system has perceptual states that ‘represent’ elements of the world and their location. • Imagination. The system can recall parts of the world or create sensations that are like parts of the world. • Attention. The system is capable of selecting which parts of the world to depict or imagine. • Planning. The system has control over sequences of states to plan actions. • Emotion. The system has affective states that evaluate planned actions and determine the ensuing action. E.g. Aleksander (2005) Example: Aleksander’s Axioms
  • 12. Consciousness for bridging a temporal gap Clark & Squire (1998) • Delay conditioning occurs without awareness • Trace conditioning requires awareness (and hippocampus)
  • 13. Online and memory guided action Perceptual judgement Visually guided action Milner et al. (1991) Memory guided action depends on perceptual system
  • 14. V1 hypothesis • V1 is not conscious because it does not project to prefrontal cortex. • The rationale from biological usefulness of awareness – A: To produce the best current interpretation of the visual scene – B: To make the information available for the system that plan and execute voluntary motor outputs. Crick & Koch (1995)
  • 15. A New Hypothesis on Consciousness A Counterfactual Information Generation Hypothesis Agents with the ability to generate predictions, including counterfactual future and past, of their own state in the environment have consciousness.
  • 17. Active inference Friston et al. (2010) Biol Cybern The motivation: Biological agents need to minimise the entropy of their sensory states to resist a tendency to disorder. (Ergodic Assumption) (Entropy of sensory states) This implies that agents should minimise surprise over time. (Average over space = Average over time)
  • 18. Active inference c.f. series of work by Friston Generative model Active inference Perceptual Inference
  • 19. Active inference with counterfactuals c.f. series of work by Friston Instead of just minimising the entropy of sensory states, more recent formulations suggest minimising the joint entropy of sensory and hidden states. We can decompose this into: (Entropy of Sensory States) + (Conditional Entropy of Hidden States)
  • 20. Counterfactual predictions Next eye positions are chosen to minimise uncertainty of hidden states
  • 21. Counterfactual predictions in a robot Bongard , Zykov & Lipson (2006)
  • 22. • Intention – Counterfactual predictions of the consequences of future actions. • Non-Reflexive Behaviour – Mental simulation detached from the present external stimuli. • Perceptual Presence (Seth 2014) – Coherent sensory motor contingency. Summary A model of the world, and sensory motor contingency are prerequisites for the ability to simulate the world internally. Function of Consciousness = The ability to represent events disconnected from the present environment.
  • 23. What does it mean to generate information? input encoding decoding output VAE: µ x σ  z Info Processing Info Generation For individual layers, the analogy represented here is not obvious. Hidden: z
  • 24. Link to predictive coding Prediction error Info. Processing Prediction Info. Generation input Hidden: z ConsciousUnconscious DecodingEncoding GenerationProcessing DecompressionCompression PredictionPrediction Error FeedbackFeedforward Feedback generates information, hence consciousness.
  • 25. Feedback and visual awareness Super et al. (2001)
  • 26. Super et al. (2001) Late activity (feedback) is associated with subjective visibility. Feedback and visual awareness
  • 27. Optogenetics evidence Manita et al. (2015) Neuron
  • 28. Creating an agent with information generation capability
  • 30. Generative control:  'What is the distribution of possible futures?' Distribution includes actions → search for actions that lead to desirable futures  Can change reward function on the fly  Reward function can easily depend on things like uncertainties and surprisals  Free long-term planning/coherency Latent Space Future Sensory State Future Action Pattern Latent Space Latent Space Latent Space Latent Variable Future Sensory State Future Action Pattern Future Sensory State Future Action Pattern Future Sensory State Future Action Pattern Future Sensory State Future Action Pattern Generative Model Generative control
  • 31. Repeated re-encoding allows an auto-encoder to better match the data distribution (Arulkumaran, Creswell, Bharath 2016) Recurrent Autoencoder
  • 32. Generative Control of Cartpole Learn action+sensor generative model, maps latent variable → prediction for next 16 frames. Gradient descent in latent space to pick action
  • 33. Ultimate method: Direct connection • Qualia seem to be shared among the brains through direct connections. • We can connect brains directly to AI to check the qualia in AI. The Hogan sisters seem to share qualia via direct connections in the thalamus.
  • 34. Final thoughts • Information generation, not processing – We should pay more attention to the generative aspect of the brain. • Qualia are not counterfactual – Perception of the present is also generated. – Corresponding bottom-up signals may help generation of more detailed, clear contents. • Intrinsic definition of information generation – IIT-like definition? • Phenomenal consciousness – If we create the causal structure of the third-person aspect of consciousness, the first-person aspect of consciousness should be generated (Biological Naturalism).