The Mechanism of Thought

Download Report

Transcript The Mechanism of Thought

Group-7
Mental World
What is Cognition?
 Cognition :
 Understanding and trying to make sense of the world
 Information processing
 Development of concepts
 The mental functions, mental processes and states of
intelligent entities with focus on:





comprehension,
inference,
decision-making,
planning and
learning etc.
Cognition
 Cerebral Cortex
 Seat for cognition
 How is Cerebral cortex able to do what it is able to do?
 Fuzzy explanations
 No clear cut perspective in exact terms
 Introducing Confabulation Theory
 A discrete insight into possible mechanism of thought
Confabulation Theory
 Proposed by Robert Hecht-Nielson
 All human cognition and behavior based on one
simple, non-algorithmic procedure that has been
named confabulation
 All aspects of cognition carried out using:
 a single type of knowledge and
 a single information processing operation called
(confabulation)
Relevance
 Radical novel approach
 First-of-its-kind concrete model
 Issuing deeper insights into process of cognition
 Augmenting present
 Artificial intelligence


knowledge discovery
knowledge management
Roadmap






Cognitive World Object Representation
Knowledge Links
Confabulation
The Mathematics of Cogent Confabulation
The Origin of Behaviour
Conclusion
Cognitive World Object Representation
 Cerebral Cortex
 4000 discrete, localized, disjoint patches
 Thalamocortical Module
 cortical patch and first-order thalamic zone uniquely
paired by reciprocal axonal interconnection
 Module
 one attribute of an object of mental world
Cognitive World Object Representation
Cognitive World Object Representation






How are these modules used?
Make groups of 60: symbols
Module may consist of thousands of neurons
Symbol: one possible descriptor of an attribute
One neuron may belong to more than one symbols
Symbols have to be permanent
Cognitive World Object Representation
What is Knowledge?
 Knowledge accumulates in discrete units
 Knowledge link or knowledge unit is an axonal linkage
between source symbol and target symbol
 Source and target symbols belong to different
modules
Knowledge Links
Formation of knowledge links
 Two symbols become co-active to send out signals
 Synapses are strengthened in the process of learning
 Permanent strengthening during sleep
 Billions of knowledge links
 Humans and animals are 'smart’
Confabulation




One and only one information processing operation
Localized
Winners-take-all
Symbol with highest total knowledge link input is
conclusion of confabulation
Implications of Knowledge Link
 Source symbol neurons send signals to millions of
transponder neurons
 Only few thousand transponder neurons become
highly excited
 10 % of the target neurons receive signals from
multiple transponder neurons
Implications of Knowledge Link
Confabulation – Neuron Level
Cogency
Cogency
 Aristotelian model : An appealing model for cognition
: p(ε|αβγδ)
 Wrong model
 Alternate model : Cogency : p(αβγδ|ε)
 α,β,γ,δ : Assumed facts
 ε : Conclusion
Cogency Theorems
 Thm 1: If αβγδ => ε exclusively, then maximization of
cogency produces one and only one answer ε
 Aristotelian logic information environment:
maximizing cogency gives logical answers
Cogency Theorems
• Cogency calculation : only in trivial situations
• Confabulation Product : p(α|ε).p(β|ε).p(γ|ε).p(δ|ε)
• Thm 2: [p(αβγδ|ε)]4 =
[p(αβγδε)/p(αε)].[p(αβγδε)/p(βε)].
[p(αβγδε)/p(γε)].[p(αβγδε)/p(δε)].
[p(α|ε).p(β|ε).p(γ|ε).p(δ|ε)]
• In non-exceptional cases,
[p(αβγδ|ε)]4≈C*[p(α|ε).p(β|ε).p(γ|ε).p(δ|ε)]
Confabulation Examples
 Some examples from test
 She could determine (whether, exactly, if, why)
 If it was not (immediately, clear, enough, true)
 Earthquake activity was [centered]
 For lack of a (unified, blockbuster, comprehensive, definitive)
 A lack of (urgency, oxygen, understanding)
 Regardless of expected [outcome, length]
 Automatic emergence of semantics and grammar
Why Cogency and not Baye’s Law?
 p(λ)=0.01; p(ε)=0.0001; p(αβγδ|λ)=0.01; p(αβγδ|ε)=0.2
 p(αβγδ|ε)= 20 * p(αβγδ|λ)
 p(λ |αβγδ)= 5 * p(ε|αβγδ)
 Baye’s Law => λ
 Cogency => ε
Quiz! (For those who are sleeping)
 Quickly select a next word for each of the following:
 Company rules forbid taking
 Mickey and Minnie were
 Capitol hill observers are
 Paper is made from
 Riding the carousel was
Why Cogency and not Baye’s Law?
 Typical answers :
 Naps
 Happy
 Wondering
 Wood
 Fun
 ‘the’ also viable : Baye’s law
Conclusion-Action Principle: Origin of
Behavior
Conclusion-Action Principle: Origin of
Behavior
 Every time a confabulation operation on a module reaches
a conclusion, an associated set of action commands are
launched
 Winner of a confabulation competition employs skill
knowledge to launch an action
 Skill knowledge, and skill learning are not parts of
cognitive thinking
Hindering Blocks
 Conceiving and then precisely defining a
confabulation architecture
 Conceiving and then precisely defining the thought
processes
 Conceiving and executing an appropriately staged
sequence of learning opportunities
Conclusion
 A new dimension to the mechanism of thought
 Based on Cogency (refutes Bayesian model)
 If proved correct
 Better insight into human cognition
 Will redefine the outlook towards various AI problems
 Radical improvement in present ML techniques
References
 Robert Hecht-Nielson, Cogent Confabulation, Neural
Networks Letter, 2004
 Robert Hecht-Nielson, Confabulation Theory: A
Synopsis, Institute for Neural Computation, 2005
 Robert Hecht-Nielson, The Mechanism of Thought,
International Joint Conference on Neural Networks,
2006
 scholarpedia.org/confabulation
ThanQ
 “Animal cognition maximizes cogency, and in a non-
logic environment, cogency maximization implements
what I call the ‘duck test.’" - Robert Hecht-Nielson
 “There must be some event that triggers every
behavioral event, and it had to be the same in every
instance, whether we're thinking, moving, or
whatever." - Robert Hecht-Nielson