Transcript Slide 1 - Department of Computer Science

Causality and Complexity in Adaptive
Neural Systems
Complex ‘2009 Workshop on “Causality in Complex Systems” (Complex CCS)
A Review-in-Progress by David Batten
CSIRO, Australia
Goal and Method
• To explore and review the concepts of causality and
complexity in brain research and cognition
• From the perspective of a complex systems scientist
only vaguely familiar with advances in neuroscience
• Making use of:
• Published papers and books in neuroscience and in related
fields (e.g. psychology, psychophysiology, etc.)
• Special issues of leading journals (e.g. the 2006 special issue
of the International Journal of Psychophysiology on the Quiet
Revolutions in Neuroscience)
• Important Conferences (e.g. the Brain Network Dynamics
Conference at UC Berkeley in honour of Walter Freeman’s
80th Birthday, 2007)
• In order to better understand, and perhaps eventually
to better model, causal and influence networks that
evolve within the human brain  human aspirations
What is Consciousness?
• According to Walter Freeman, the pertinent questions
are:
• How and in what senses does consciousness cause the
functions of our brains and bodies?
• How do brain and body functions cause consciousness?
• How do actions cause perceptions?
• How do perceptions cause awareness?
• How do states of awareness cause actions?
• Analysis of causality is a necessary step towards a
better comprehension of consciousness
• The types of answers depend on the choice among
meanings that are assigned to the word “cause”:
• linear causality
• circular causality
• non-causal interrelationships
Linear Causality of the Observer
Source: Walter Freeman (1999)
Linear Causality in Action
• A stimulus Sn initiates a chain of events including
•
•
•
•
•
•
Activation of receptors
Transmission by serial synapses to cortex
Integration with memory
Selection of a motor pattern
Descending transmission to motor neurons
Activation of muscles
• At nodes along the chain, awareness occurs, and
meaning and emotion are attached to the response
• Temporal sequencing is crucial; no effect can precede
or occur simultaneously with its cause
• At some instant, each effect becomes a cause
• This conceptualization is inherently limited, because
awareness cannot be defined at a point in time.
Circular Causality of the Self
real
time
death
Source: Walter Freeman (1999)
Circular Causality in Action
• The double dot shows a point moving counterclockwise
on a trajectory idealized as a circle, showing that an
event exists as a state through a period of inner time,
which we reduce to a point in real time.
• Stimuli from the outside world impinge on this state.
• So also do stimuli arising from the self-organizing,
interactive dynamics within the brain.
• Most stimuli are ineffective, but occasionally one does
succeed as a "hit" on the brain state, and a response
occurs.
• The impact and motor action are followed by a change
in brain structure that begins a new orbit.
• So, changing our (state of) mind changes the neural
structure of our brains
Circular Causality = Systemic Causality?
• A succession of orbits can be conceived as a cylinder
with its axis in real time, extending from birth to death in
an individual and its brain
• Trajectories in inner time may be viewed as fusing past
and future into an extended present by way of state
transitions
• Circular causality expresses the interrelations between
levels in a hierarchy
• A top-down macroscopic state simultaneously influences its
microscopic elements, and
• The microscopic elements create and sustain the macroscopic
state from the bottom up
• The circular and hierarchical relationship between such
microscopic and macroscopic entities is essential for
explaining brains; also lasers (see Haken, 1983).
Some of Freeman’s Conclusions
• Awareness cannot be explained by linear causality
• Intentionality cannot be explained by linear causality
• Interactions between microscopic and macroscopic domains
of the brain accord with the laws of self-organization
• Circular causality in a self-organizing brain is a concept
that is useful to describe interactions between microscopic
neurons in assemblies and the macroscopic emergent state
variable that organizes them.
• New methods are needed to explain how all those neurons
simultaneously get together in a virtual instant & switch from
one harmonious pattern to another in an orderly dance!
• A surprisingly similar pattern switching holds for:
• the excitation of atoms in a laser to produce light (Haken)
• the metamorphosis of caterpillars into butterflies
• the inflammatory spread of epidemics or behavioural fads
New Method 1: S-O and Synergetics
• Synergetics and self-organization of brain function and
cognition (Haken, Kelso, Freeman, Lewis)
• Circular causality describes bidirectional causation between
different levels of a system (Haken, 1977). Maurice MerleauPonty introduced the concept, claiming that every action and
every sensation is both a cause and an effect.
• Brain dynamics is governed by an adaptive order parameter
that regulates everywhere neocortical mean neural firing rates
at the microscopic level, finding expression in the maintenance
of a global state of self-organized criticality (Freeman, 2004)
• The concept of circular causality should be discarded
(Bakker)
• Circular causality suggests an interaction between separable
entities that does not exist.
• The micro-macro relationship is one of correspondence rather
than causation
New Method 2 – Attractor Neural Networks
• Hopfield introduced the general concept of an attractor
neural network (ANN)
• In his 1982 paper on neural networks as physical systems
with emergent computational abilities, he defined an
associative memory model based on formal neurons
 the first mathematical formalisation of Hebb’s ideas and
proposals on the neural assembly, the learning rule, the role
of connectivity in the assembly and the neural dynamics.
• ANNs are being used to confirm the hypothesis that a
collective phenomenon is at the origin of our memory
function (Amit and others).
• Important associated concepts are:
• Synaptic plasticity – based on Hebbian rules
• Continuous ANNs
New Method 3: Causal Networks
• Neurons engage in causal interactions with one another
(self-organization) and with the surrounding body and
environment (adaptation)
• Neural systems can thus be analyzed in terms of causal
networks, without assumptions about info processing;
• e.g. using Granger causality & graph theory
• A neurobiotic model of the hippocampus & surrounding
area identified shifting causal pathways during learning
of a spatial navigation task:
• Selection of specific causal pathways – “causal cores”
• Causal network approach may help to characterise the
complex neural dynamics underlying consciousness:
• Causal density as a candidate measure of neural complexity
• The Neurosciences Institute – Seth, Edelman, Tononi
Distinguishing Causal Interactions (Seth)
Granger Causality
• Clive Granger – Nobel prizewinner in economics for his
work in econometrics on time-series analysis
• Granger causality is a method for determining whether
one time series is useful in forecasting another
• Ordinarily, regressions reflect "mere" correlations, but Granger
argued that there is an interpretation of a set of tests that can
reveal something useful about causality.
• Statistical, not physical
• Causality can be unidirectional or reciprocal
• Many extensions to suit neurodynamics:
• e.g. Multivariate Granger causality
• e.g. Nonlinear Granger causality
• Granger causality interactions can
be represented as a directed graph
Lakoff on Frames and Metaphors
• “Frames” are mental models of limited scope
• e.g. our traditional frame for war includes semantic roles like
nations at war, leaders, armies with soldiers and commanders,
weapons, attacks, battlefields, etc.
• Such frames + metaphors (e.g. “nerves of steel”) in our
brain define our “common sense”
• Human thinking in frames and metaphors gives rise to
inferences that don’t fit the laws of logic or deductive
rationality as e.g. economists have formulated them
• Because facts matter, undistorted framing is needed to
communicate the truth about our economic, social and
political realities
• Differing worldviews or aspirations often lead to the
proliferation of distorted frames and metaphors
Two Competing Worldviews
• There may be as many worldviews as human beings?
• In the social sciences, a few worldviews crop up time
and again:
•
•
•
•
Sheep and Explorers (in traffic)
Imitators and Innovators (in technology)
Cartesians and Stochasts (in fishing strategies)
Conservatives and Progressives (in politics)
• They correspond to 2 extremes in terms of risk-taking
behaviour or creativity
• Lakoff: 2 parenting models  2 worldviews
• Strict father model  Conservatives  Linear Causality
• Nurturant parent model  Progressives  Systemic Causality
• Many people retain active versions of both models in
different parts of their brain, and use them in different
parts of their lives
Conclusions for our Workshop series
• Causality and complexity have been discussed at length
by scholars in the field of neuroscience
• especially linear versus circular circularity
• especially with respect to neural nets and causal networks
• Thus it could be worth focusing on neuroscience as a
subtheme at one of our workshops
• At the forefront of causality discussions have been:
•
•
•
•
•
Walter Freeman, UC Berkeley
Hermann Haken, U of Stuttgart
Anil Seth, U of Sussex
Steve Bressler, Florida Atlantic U
Several scholars at The Neurosciences Institute, San Diego
• Several others could be worth our attention:
• e.g. George Lakoff, UC Berkeley
Thank you
Dr. David Batten
CSIRO, Australia
Phone:
Email:
+61 3 9239 4420
[email protected]
Thank you!
Contact Us
Phone: 1300 363 400 or +61 3 9545 2176
Email: [email protected] Web: www.csiro.au
Three Worldviews
• Individualism
• Reduce all social constructs to collections of individuals (micro,
no emergence)
“There is no such thing as society” – Thatcher
• Holism
• Structure dominates composition (macro, no emergence)
“Any society does not consist of individuals but expresses the
sum of relationships [and] conditions that the individual actor
is forming” – Marx
• Systemism
• Model entities by composition, environment, structure and
mechanism (micro and micro, emergence)
“Systemism makes room for both agency and structure”
– Bugne
Source: Alex Ryan (2007)
What is a System?
• Interdisciplinary concept with 2 core influences:
• Emergence and Hierarchy (General Systems Theory)
• Communication and Control (Cybernetics)
Contemporary Systems Approaches