Transcript The Brain Doesn`t Work That Way: From Microgenesis to Cognition

The Microgenetic Dynamics of
Cortical Attractor Landscapes
Mark H. Bickhard
Lehigh University
[email protected] http://bickhard.ws/
Abstract
• Attractor landscapes are dispositional
models of neural processes, but those
landscapes themselves have a dynamics. I
will outline how such landscapes are
ongoingly created and modified, and how
primitive representation emerges from these
processes.
Context:
The Broader Model
• Ontological Emergence
• Conceptual barriers from Pre-Socratics
– Hume
– Kim
• Emergence of Normativity
• Also ancient problems
– Biological function
– Representation
Representation
• Cognition and Representation emerge
in interaction systems
– Self-maintenant systems
– Recursively self-maintenant systems
• Selection of interaction = presupposition of
appropriateness; anticipation of
appropriateness
– ‘Appropriateness’ is normative
– Derives from underlying model of normative function
• Yields truth value — representation
Pragmatism
• An interaction based, pragmatic, model
of representation
– Kinship to Piaget
• More complex representations
– Objects
– Abstractions: e.g., numbers
Interaction Requires Timing
• Successful interaction requires timing
coordination
– This is coordinative, neither too fast nor too
slow
• Turing machines cannot handle timing
• Computers have central clocks
– Not plausible for the brain
Timing Requires Oscillators
• Solution: Put clocks everywhere
• But clocks are “just” oscillators
– Functional relationships are relationships among
oscillators: modulations
– Trivially at least TM powerful
• Need a tool kit of different forms and scales of
modulation
– Modulations of modulations … of oscillatory
activity
And This is What We Find
• Neurons are standardly modeled as:
– Threshold switches
– Connectionist nodes
– Frequency encoders
• All have in common the assumption that
neurons are ‘just’ input processors
• And that neurons are the only functional units
Both Are Wrong
• Neurons and neural circuits are endogenously
active
– In multiple ways
– They do not just process inputs
• And neurons are not the only functional
units
– Glia, for example, are also functional, not
just supportive
Neurons
And local circuits
• Oscillators
– Resonators
• Multiple interesting implications
– Modulations of endogenous activity, not
switches of otherwise inert units
Neurons II
•
•
•
•
Silent neurons
Interneurons
Short connections
Volume transmitters
• L-Dopa
• Graded release of transmitters
• Gap junctions
• Why multiple transmitters if all synapses are
classical?
• Transmitters evolved from hormones
• Classical synapses evolved from volume transmitters
Astrocytes (Glia)
•
•
•
•
•
•
•
Receive transmitters
Emit transmitters
Form functional “bubbles”
Gap junction connections
Calcium waves
Modulate synaptogenesis
Modulate synaptic functioning
– Release, uptake, degree of volume diffusion, …
Confirmation of Implication of
Model of Representation
• So, we do find a rich toolbox of multiple
scales of modulatory relations
Now In Reverse
• CNS functioning implies anticipatory
cognition
Multiple Scales
• These are all modulatory influences at
multiple scales
– Large and small spatial scales
– Slow and fast temporal scales
– There are also variations in delay times
• Evolution has created a large tool box of
multiple kinds and scales of modulatory
influences
Microgenesis:
Large Temporal Scale
• Larger and slower processes set the context
for smaller and faster processes
• They set the parameters for the faster and
smaller processes
– Ion and transmitter concentrations
– Modes of synaptic functioning
• They generate vast concurrent micro-(and
meso-) modes of processing across the brain:
Microgenesis
Dynamic Programming
• Parameter setting for dynamic
processes is the dynamic equivalent of
programming in a discrete system
• Microgenesis sets and changes the
programs across the brain
• Microgenesis is ongoing and occurs in
real time
Functional Anticipation
• Microgenetic set-up may or may not be
appropriate to the actual flow of
interactive processing that occurs in the
organism
• Microgenesis is functionally anticipatory
– The anticipation is that the microgenetic
set-up will be appropriate
Emergence of Truth Value
• Microgenetic anticipations can be true
or false
– And can be functionally determined to be
false if the interaction violates anticipations
• This is the emergence of
representational truth value out of
pragmatic functional success and failure
Content
• Microgenetic anticipations will be true in some
environmental conditions, and false in others
• Microgenetic anticipations, then, presuppose that the
appropriate conditions — whatever they are — obtain
in the current environment.
– The flow of anticipated conditions is implicit in the flow of
microgenesis
• Those conditions constitute the content of the
representing
– An implicit content
How Does This Differ?
• Endogenously active
• Interaction based, not input processing
• Future oriented, not past oriented “spectator”
model (Dewey)
• Inherently modal: anticipations of interaction
possibilities, not foundationally built on
encoding correspondences with actual
particulars
• Implicit, thus unbounded, not explicit
– Frame problems
• Etc.
Two Way Implication
• So, analysis of representation yields a
required substrate of multi-scale modulatory,
interactive brain processes
• And an oscillatory/modulatory tool kit is
precisely what we find
• And, analysis of how the brain functions
yields an anticipatory, interactive model of
representation
• Each implies the other
Microgenesis:
Larger Spatial Scale —
Attractor Landscapes
• The slower scale processes engage in
microgenetic programming of faster
processes
• The larger scale of these processes —
astrocytes, volume transmitters, short range connections, reciprocal
— induces weak
coupling among oscillatory processes
• Such weak coupling induces attractor
landscapes
connections with thalamus, etc.
– Within which faster processes proceed
Modulation of Attractor
Landscapes
• Modulation of microgenesis, therefore,
modulates attractor landscapes
 \ Modulation of slower, larger scale process
— astrocytes, etc. — modulates attractor
landscapes
• Provides a new framework for interpreting
functionality of prefrontal - basal ganglia thalamus - cortex loops
– As engaged in modulation of attractor landscapes
Thought
• These loops generate a kind of internal
interaction with the dynamic spaces
within which other CNS processes take
place
• This fits well with Pragmatic/Piagetian
conception of thought as internal
(inter)action
Further Issues
• Other models of representation
– Millikan
– Dretske
– Fodor
– Cummins
– Encodingism
Further Issues II
• Other phenomena of mind
•
•
•
•
•
•
•
•
•
•
•
Perception
Memory
Motivation
Learning
Emotions
Reflective consciousness
Language
Rationality
Social ontology
Personality, psychopathology
Ethics
Conclusion
• In being intrinsically interactive,
representation and cognition are
inherently:
•
•
•
•
•
•
Future oriented, anticipative
Pragmatic
Modal
Situated
Embodied
…
Conclusion II
• And they are realized in:
– Internal interactive modulations of
– Attractor landscapes for
– Oscillatory/ modulatory control of
– Interactions of organism with environment
Fini
What’s Wrong with Standard
Models of Representation?
• Encodingism
– Error, system detectable error — radical
skeptical argument
– Which correspondence?
– Copy argument — Piaget
– Externally related content: regress of
interpreters
– Partial recognition of problems: empty
symbol problem, grounding problem
What’s Wrong With Standard
Models? II
• Millikan
– Representation as function
– Etiological function is causally epiphenomena
• Dretske
– Etiological function again, learning history rather
than evolutionary history
• Fodor
– Asymmetrically dependent counterfactual relations
• Counter example of crank molecule
What’s Wrong With Standard
Models? III
• Error
– From observer perspective
• Millikan OK
• Dretske OK
• Fodor Sort of OK
• System detectable error
– Content is not system accessible for any of these
models
– Comparing content with what is supposed to be
being represented to determine truth or error is
representational problem all over again
– They are circular with respect to this criterion
What’s Wrong With Standard
Models? IV
• Symbol system hypothesis
– Transduced encoding
• Connectionism
– Trained encoding
What’s Wrong With Standard
Models? V
• Dynamic systems
• The interactive model is clearly a
dynamic, process model
• Dynamic approaches, however, are
often anti-representational
– E.g., Van Gelder, Thelen
Dynamic Systems Approaches
• But, dynamic systems as agents must select
interactions,
 \ must functionally indicate interaction
potentialities,
 \ must yield representational truth value
 \ must involve normative representation, whether
that terminology is used or not
– Criticisms of representation are in fact criticisms of
encodingist approaches to representation
Encodingism
• Encodings do exist
– But they borrow content
– E.g., Morse code
– They cannot generate emergent content
• Serious problem for learning
• E.g., Fodor’s innatism
• Encodingism assumes that all representation
is of encoding form
• Encodingism does not work
Further Issues
• Contemporary work pervasively assumes
encodingism:
–
–
–
–
–
–
–
–
Perception
Rationality
Language
Memory
Learning
Emotions
Consciousness
…
Conclusion I
• Representation is interactive, future
oriented, pragmatic, non-encoding,
modal, situated, embodied, and so on.
Conclusion II
• These force multiple further changes:
– Perception
– Language
– Memory
– Motivation
– Learning
– Models of Brain Processes
– And so on
Conclusion III
• A major reworking of our models of and
approaches to the whole person is
required
– The Whole Person