Ecological Information Makes
Embodiment Possible
Andrew D Wilson, Sabrina Golonka
Leeds Beckett University
@PsychScientists
http://psychsciencenotes.blogspot.co.uk/
So you want to do embodied cognition…
Embodied cognitive science
• Goal is to solve the grounding
problem of representations by
making them embodied
Radical embodied cognitive
science
• Explains behaviour using different
tools with consequence that
representations never show up
Chemero 2009
Embodied Cognition is not What You Think It Is
Wilson & Golonka (2013)
“…if perception-action couplings and resources distributed over brain,
body, and environment are substantial participants in cognition, then
the need for the specific objects and processes of standard cognitive
psychology (concepts, internally represented competence, and
knowledge) goes away, to be replaced by very different objects and
processes…”
Wilson AD and Golonka S (2013) Embodied cognition is not what you
think it is. Front. Psychology 4:58. doi: 10.3389/fpsyg.2013.00058
https://xkcd.com/927/
We will
replace
you if we
win
Embodied Cognition is not What You Think It Is
Wilson & Golonka (2013)
“…to be replaced by very different objects and processes…”
Dynamical mechanistic explanations held
together by ecological information
Outline
Dynamical mechanistic explanations (Bechtel and others)
• What they are and why we might want aspire to them
Explaining behaviour with RECS (Wilson & Golonka, 2013)
• Identify your task; Identify your resources; Propose a model using those
resources that produces that behaviour; Science the model
Worked Example
• Coordinated rhythmic movement (Bingham, 2001, 2004a, b; Golonka &
Wilson, 2012)
Ecological information enables mechanistic explanation in embodied
cognition
Dynamical Mechanistic Explanation
Bechtel & Abrahamsen, 2010
A mechanism is a structure performing a function in
virtue of its component parts, component operations
and their organisation. The orchestrated functioning of
the mechanism, manifested in patterns of change over
time in properties of its parts and operations, is
responsible for one or more phenomena
Dynamical Mechanistic Explanation
Bechtel & Abrahamsen, 2010
Two strategies for modelling mechanisms
1) Advance de novo proposals for mechanism; evaluate by showing
this model produces the phenomenon of interest
Relationship between model and actual system remains unknown
2) Formally describe mechanism made of parts independently
established to be in that mechanism; use to probe the
phenomenon of interest
Expands the range of ways to investigate a system
Dynamical Mechanistic Explanation
Bechtel & Abrahamsen, 2010
Two strategies for modelling mechanisms
1) Advance de novo proposals for mechanism; evaluate by showing
this model produces the phenomenon of interest
Cognitive Science (e.g. neural networks)
2) Formally describe mechanism made of parts independently
established to be in that mechanism; use to probe the
phenomenon of interest
Biology (e.g. circadian rhythms)
Dynamical Mechanistic Explanation
Bechtel & Abrahamsen, 2010
Because models in cognitive science typically don’t
involve independent evidence for the involvement of
their constituent parts, they struggle to provide the
dynamical mechanistic explanation of a system’s
behaviour that the other models can
Dynamical Mechanistic Explanation
Wilson, just now
Because of degeneracy (many:one structure to function
mapping) and neural reuse (one:many) this modelling
strategy is doomed and cannot provide evidence about
the actual mechanism
Dynamical Mechanistic Explanation
Silberstein & Chemero, 2013
All the good stuff in cognitive science is nondecomposable so mechanistic analyses fail
Systems neuroscience, graphs, networks, etc
Conflates ‘this model is handy’ with ‘this model
explains’
A Research Programme
Empirically identify the component parts and
operations of the mechanism
THEN
Model it so you can poke around, make
predictions etc
Wilson & Golonka (2013)
1. What is the task to be solved?
2. What are the resources that the organism has access to in
order to solve the task?
3. How can these resources be assembled so as to solve the
task? (Modelling time!)
4. Does the organism, in fact, assemble, and use these
resources?
Wilson AD and Golonka S (2013) Embodied cognition is not what you think it is. Front. Psychology
4:58. doi: 10.3389/fpsyg.2013.00058
Tasks All the Way Down
Behaviour is task-specific
• It has to be, or else we would never be able to pick one
behaviour from the huge number available (Bingham, 1988)
• It has to be, or else how can evolution and learning evaluate
functionality?
• It has to be in REC because we model behaviour as emerging
from the real time activity of an embedded agent (i.e.
embedded in a task)
So what’s a task?
Task Dynamics
Events in the world unfold over time in a particular way
This pattern is the result of the composition and
organisation of the underlying task dynamic
Events are only uniquely identifiable at the level of
dynamics (Bingham, 1995)
Event in the world = bouncing ball (with friction)
Dynamical
description =
damped mass spring
Two tasks are the same if the underlying
dynamic is the same (even if the
parameters are different)
Two tasks are different if they have a
different underlying dynamic
In order to act appropriately in a given task, we need to
identify the task we are facing
We therefore need to perceive the underlying task
dynamic
‘‘The central question for the
theory of affordances is not
whether they exist and are real, but
whether information is available in
the ambient light for [directly]
perceiving them’’
(Gibson, 1979, p.140)
The Perceptual Bottleneck
Dynamics (in the world)
creates
Kinematics (information)
Kinematic Specification of Dynamics
We can only perceive the relevant dynamics if those
dynamics have kinematic consequences that we can
detect
Kinematics can specify dynamics (Gibson, 1979; Turvey
et al, 1981; Runeson & Frykholm, 1983)
BioMotion Lab Walker
All We Have Is Information
REC: Behaviour emerges in a context: a task
• Behaviour is functional if it is appropriate with respect to
that task
• Tasks are only distinguishable from each other at the level of
dynamics
SO
• Organisms need to perceive dynamics to select functional
behaviour
BUT
• No direct access to dynamics
BUT LUCKILY
• each dynamic lawfully creates different kinematic patterns
in energy arrays which can specify those dynamics
SO
• Organisms can detect these patterns and use them
• as if they were the relevant dynamics (this works because of
specification; Turvey et al, 1981)
• as if they were about something like language (this works
because enough people agree; convention, Golonka, 2015)
Patterns used this way become ecological information
If you select, coordinate and control behaviour
with respect to task specific information then that
behaviour will by definition be functional (task
appropriate) and require no ‘mental gymnastics’
Ecological information makes radical embodied
cognition possible
Wilson & Golonka (2013)
1. What is the task to be solved?
2. What are the resources that the organism has access to in
order to solve the task?
3. How can these resources be assembled so as to solve the
task? (Modelling time!)
4. Does the organism, in fact, assemble, and use these
resources?
Geoffrey P Bingham
Indiana University
Coordinated Rhythmic
Movement
0° & 180° the only stable
states
0° more stable than 180°
V = -a cosφ –b cos 2φ
Other coordinations (e.g. 90°)
can but must be learned
1. Identify the Task
To produce this dynamic, experiments reveal that you
will need:
• Two things moving rhythmically, in parallel and
coupled to each other (φ uniquely defined)
• The two things can belong to different people, but
at least one thing must belong to a person (the
coupling is not neural, but perceptual)
http://psychsciencenotes.blogspot.co.uk/2010/11/perceptionaction-model-of-coordination.html
2. Identify Resources
Two oscillators producing parallel harmonic motion so
as to move at some relative phase to each other
Information about that motion and that relative phase
to couple the oscillators and allow error correction to
maintain that relative phase
2. The Action Resources
Key results:
• Limit cycle stability (overall
dynamic is non-linear)
• Phase resetting (overall
dynamic must be
autonomous)
• Inverse
frequency/amplitude
relation (↑ stiffness)
Kay, Kelso, Saltzman & Schöner, 1987; Kay, Saltzman & Kelso, 1991
2. The Information Resources
The oscillators must be coupled by a variable that
• specifies relative phase
• is composed only of state variables (to preserve
autonomy)
• is most readily detected at 0°, then at 180°, not at
all at 90° (basic phenomena)
• detection varies with frequency
2. The Information Resources
The information should
• be detectable by both vision (e.g. Bingham et al 1999,
2001; Zaal et al, 2001) and proprioception (Wilson et al,
2003)
• only be defined for parallel movements (Bogaerts et al,
2003; Wilson et al, 2005; Wimmers et al, 1992)
• be available throughout the trajectory but vary with
relative speed (Bingham, 2004)
Information
Perceived relative
phase (rho)
Relative Speed
(noise)
Ρ = sgn(sin(Φ1) sin(Φ2) + α(Vi – Vj)3 Nt )
Relative Direction
(information)
Relative direction (kinematic information)
specifies relative phase (dynamical property)
0°: relative direction always the same
180°: relative direction always different
90°: relative direction maximally variable
3. Model Built From Those Resources
.
x¨ 1 + b x1 + kx1 = c sin (Φ2) Ρij
.
x¨ 2 + b x2 + kx2 = c sin (Φ1) Ρji
where:
Φ = arctan (Vn / x)
Ρ = sgn(sin(Φ1) sin(Φ2) + α(Vi – Vj)3 Nt )
Bingham (2001, 2004a, b)
4. Science!
4. Science!
4. Science!
Unexpected
discovery: not
everyone uses
relative direction at
0° and 180°!
4. Science!
• Explains why effects persist between people
• Explains why effects persist without movements
• Explains why effects go away if the feedback
removes relative direction (and motivated new
feedback)
• Explains why learning rates vary (faster
close to 0°)
4. Science!
• Successfully predicted perceptual learning would
improve movements (Wilson et al, 2010)
• Successfully predicted transfer of learning between
unimanual and bimanual versions (Snapp-Childs et al,
2015)
• Successfully predicted relative speed is relevant
noise term (Snapp-Childs et al, 2011)
Replacement via Winning
HKB Dynamic Patterns Approach
Ecological Information Approach
• 0° and 180° are attractors,
therefore behaviour is stable
there
• Attractors make learning harder
• Φ defined with respect to
muscles
• Behaviour organised with
respect to Φ
• Information & therefore
behaviour is stable there,
therefore they are attractors
• Stable info makes learning easier
• Effects persist between people,
in judgments, etc
• Behaviour organised with
respect to information
Dynamical Mechanistic Explanation in REC
Formally describe mechanism made of parts independently established
to be in that mechanism; use to probe the phenomenon of interest
Bechtel & Abrahamsen, 2010
1. What is the task to be solved?
2. What are the resources that the organism has access to in order to
solve the task?
3. How can these resources be assembled so as to solve the task?
(Modelling time!)
4. Does the organism, in fact, assemble, and use these resources?
Wilson & Golonka, 2013
Coupling & Constitution
Coupling
The Body
The World
creates
Harmonically
oscillating limbs
Information
Mechanism
couples
The elements of this mechanism constitute coordinated rhythmic movement
Tasks are identified at level of dynamics
Task dynamics creates information
Information specifies but is not identical to dynamics
Behaviour is organised with respect to information
• Information stands in for dynamic that created it: laws (Turvey et al 1981)
• Information stands in for something else: convention (Golonka, 2015)
Behaviour is therefore functional with respect to the task
dynamics
Ecological information makes embodiment
possible
“COST is a unique means for European researchers, engineers and
scholars to jointly develop their own ideas and new initiatives across all
fields of science and technology through trans-European networking of
nationally funded research activities.”
Applying to start a COST action dedicated to developing an ecological
approach to neuroscience
Come talk to me or email Sabrina (s.golonka@leedsbeckett.ac.uk) if
you’d like to discuss
Thanks!
Andrew D. Wilson
School of Social, Psychological, &
Communication Sciences
Leeds Beckett University
https://cognitioninaction.wordpress.com/
Notes from Two Scientific Psychologists
http://psychsciencenotes.blogspot.co.uk/
Twitter: @PsychScientists