psychology essay
Minfu Liang
05_EmbodiedCognition.pdf
10/16/2019
1
Embodied Cognition
Outfielder Problem
Outfielder Problem
• How do we know objects are coming directly towards us? • They get larger! Visual angle increases.
• How do we know where something is going when it is moving? • For linear motion – directly perceived? (representational
momentum) • For parabolic motion - calculate parabolic trajectory of object?
• But baseballs are small and are moving REALLY fast. Do we have enough time to compute parabolic movement and react? • We could predict where the ball will be due to its parabolic motion… • But again, it’s really small and really fast, and even a small initial
“miscalculation” would get us to the wrong place.
• So how can outfielders do this?
10/16/2019
2
Outfielder’s Solution? • When the outfielder moves, the way
the ball moves changes based on their own movement • Can constantly update target position
based on relative motion of ball and body
• Two possibilities for this • Optical acceleration canceling (OAT)
• Linear optical trajectory (LOT)
• But is this what they are really doing? • If predicting based on parabolic
calculation – should run straight to target location
• If updating based on body movement and ball movement comparison – should run in parabolic curved way (mirrors ball movement) Shapiro, 2013
Why? • Fielder moves laterally in order to make ball move
in straight line (relative to body movement; LOT) • Hence, mirroring the movement of the ball • LOT has more support than OAT • Baseball players likely use both LOT and OAT depending
on situation and what properties of movement are most easily visible
• Linear movement can be perceived directly • We don’t need to have complex calculations and
representation of linear movement – prediction is very simple moment-to-moment
• Goal – take complex mental operation and utilize ALL resources (including bodily resources) to turn problem into one solved by direct perception
10/16/2019
3
Perception and Action
• Implication – we don’t need to have complex abstract mental representation of movement and physics
• Actions can be planned moment to moment based on the body and desired goal
• Perception-action loop
• Processing isn’t localized in brain! Environment and body are PART of cognition
Example: How do we localize sound?
• Binaural cues (both ears)
• Interaural Time Difference (ITD) – sound reaching opposite ear is delayed
• Interaural Level Difference (ILD) – sound reaching opposite ear is quieter
• Monaural cues • Pinna folds – shape of ear
• Shape of ear varies across species – need to represent own body to utilize this information
10/16/2019
4
Cone of Confusion
• Elevation and 3d make things a bit trickier
• Sometimes ITD and ILD are ambiguous
• Pinna cues can help with elevation, but still a large amount of ambiguity
Embodied Solution!
• Considering that organisms can adjust their own perceptual input (through moving)…
• The ambiguity raised by the CoC only lasts for as long as the organism is 100% stationary, with no movement or head tilt
• Not an obvious solution from looking at envatted brain model
Determining Depth…again
• Ambiguity in depth is only ambiguous for certain angles – once you move, depth ambiguity is diminished
• Is the inverse optics problem really a problem? • We have another “dimension”
available to us – how things change over time based on the movement of our eyes, head, and body!
• This disambiguates the 2d to 3d problem
• If our perceptual systems confront ambiguity, we can use our bodies to resolve this
10/16/2019
5
Accidental Viewpoints
10/16/2019
6
Physiological Depth Cues
• Remember accommodation and convergence? • Shape of lens changes to focus light most sharply on the
retina for an object at a given distance
• This process itself provides a lot of depth information without any mental calculation!
• Gibson – shape of lens “resonates” with maximal acuity
What is this resonance thing exactly? • Attractor basin
• Preferred stable state of network (usually approaching low energy state)
• Error correction (in two dimensions) • Make a change in one direction (too far)
• Make a change in opposite direction related to degree of error (too far, but closer)
• Make a change in original direction related to degree of new error…etc.
• Eventually, reach state of minimal error (attractor basin)
10/16/2019
7
Resonance and accommodation
• Attractor basin – maximum edge clarity (sharpness)
• Directions of movement – tighten or loosen ciliary muscles • Adjust in opposite direction (tight/loose) of last
“overshoot” adjustment
• Assuming maximal sharpness is attractor on its own, process is “resonant” and not under higher level control • Just like a ball placed on a slope will roll downhill due to
gravity (you don’t need someone pushing it down)
Depth over Time • Motion parallax is also an
incredibly informative depth cue, and can be directly perceived • Even if our whole body isn’t
moving, all we need to do to get parallax information is slightly move our head
• Also provides lots of depth info without calculation – direct perception
• Emulating this on flat screens can be as striking as emulating binocular rivalry!
https://www.youtube.com/watch?v=iwrwc1BiHrg
10/16/2019
8
So what is the basic idea behind embodied
cognition? Our bodies and the richness of our environments allow for complex behavior without appealing to
high-level abstract symbolic processing
Six Views of Embodied Cognition
1. Cognition is situated. • Cognitive activity takes place in the real world • Perception and action are inherent in cognition
2. Cognition is time-pressured. • Since cognition is situated, we need to focus on how it works
under the time-constraints inherent in the environment it is present in
3. We off-load cognitive work onto the environment. • Because our internal cognitive processes (information
processing abilities) are restricted (e.g., working memory capacity), we can use the environment to hold and even manipulate information for us (e.g., writing a list down onto paper)
Wilson, 2002
Six Views of Embodied Cognition 4. The environment is part of the cognitive system
(cognition is embedded). • As the flow of information between mind and environment is
dense and uninterrupted, studying mental cognition while discounting the environment is not meaningful
5. Cognition is for action (cognition is enactive). • The function of cognition (and the mind in general) is to guide
action, and thus there needs to be a focus on the production of action itself (e.g., how does LTM recall relate to action)
6. Off-line cognition is body based. • Even when attempting to de-couple cognition from the
environment (as is the case in many cognitive research programs), the systems studied evolved to utilize the environment and are grounded in sensorimotor processes
Wilson, 2002
10/16/2019
9
Embodied Cognition
• Given the six views prior, the way that we think about cognition needs to change
• Shift from envatted brain to embodied brain
• Does this mean we need to change some of those basic research assumptions we mentioned earlier? • Ecological validity of cognitive research programs?
• How can we design experiments to be ecologically valid under an embodied cognition framework?
Four “Key Questions” for designing EC Experiments
1. What is the task to be solved? • Clearly define what the task is, being very specific about focusing on
this particular task - not just similar tasks
2. What are the resources the organism has access to in order to solve the task? • Cognitive, bodily, and environmental resources are all accessible to the
given organism for a given task • We should focus on all available resources
3. How can these resources be assembled to solve the task? • Determine the “best” way to solve the given task using the given
resources • “Best” might prioritize accuracy, latency, effort, etc. depending on the
perceived task demands
4. Does the organism, in fact, assemble and use these resources? • An empirical question – this is what the experiment should be
examining Wilson & Golonka, 2013
Outfielder Problem 1. What is the task to be solved?
• Catch the baseball/frisbee
2. What are the resources the organism has access to in order to solve the task? • Vision (specifically determining size of object through visual angle
and motion), ability to move own body • Note that we should NOT assume organism has representational
computational physics cognition (which we don’t explicitly know) – we only need to say what we KNOW the organism has (above)
3. How can these resources be assembled to solve the task? • Move body in such a way that visual angle of baseball is always
increasing (means you are getting closer to it) • Move body in such a way that ball movement is linear (LOT)
4. Does the organism, in fact, assemble and use these resources? • Looking at videos - yes
10/16/2019
10
Back to counting
Counting as EC Research 1. What is the task to be solved?
• “Add five plus three”
2. What are the resources the organism has access to in order to solve the task? • Vision, ability to count discrete visual objects, fingers
(discrete visual objects), ability to curl/uncurl fingers
3. How can these resources be assembled to solve the task? • Curl all fingers; uncurl three fingers on one hand using
visual/motor counting; uncurl five fingers on the other hand in the same way; visually count how many fingers are uncurled
4. Does the organism, in fact, assemble and use these resources? • Looking at video - yes
Is this the ONLY way to count?
• Clearly, I can do addition without adding on my fingers.
• How can we explain adding 171 + 322 in this same fashion?
• Do we now need to assume propositional representation?
• How could we represent those numbers, and do the addition, in analog?
• We’ll come back to this later.
10/16/2019
11
Recap on a finding
• We talked about this the first day of class… • Participants read words off of a
computer screen • Instructions – “respond as quickly
and as accurately as possible” • IV – either the first letter is
primed or not (using some sort of priming method, whatever kind)
• DV – reaction time (how long it takes to start reading the word)
• Results – participants are faster at reading words when the first letter has been primed
_____ m____
math math
Longer RT Shorter RT
Alternative Explanation • Let’s go to the four “key questions” for embodied research,
and evaluate this study.
1. What is the task to be solved? • Speak the word as quickly as possible, while also speaking the
correct word (be accurate)
2. What are the resources the organism has access to in order to solve the task? • In one condition, the person knows the first letter of the word • The person has articulators with which to speak the word
3. How can these resources be assembled to solve the task? • The first letter of the word is sufficient to begin speaking the word
(i.e., moving the mouth or even elongating the first segment) • The earlier speech begins, the better the solution for the task
4. Does the organism, in fact, assemble and use these resources? • Do people move their mouths to the target position based on the
initial letter prime alone? • Or even more extreme – do people begin speaking the target word
based on the initial letter prime alone?
Results
• Do people begin to move their mouth into the target position based solely on an initial segment (letter) prime? • YES
Holbrook, Kawamoto, & Liu, 2018
10/16/2019
12
Results
• Going a step further – do people actually begin to speak the word based solely on an initial segment prime? • YES!
Re-evaluating the Model
• What does all of this mean for modeling speech production?
• Levelt model – Perception -> Representation -> Action • Prime allows for encoding to occur before unprimed situation • Encoding (a process involving representation) occurs faster,
thus shorter RT • All explained through “envatted” processes
• Alternative model – Perception -> Action • People begin action based on perceived prime (moving
mouth, or even producing speech and elongating first letter - “mmmmmap”)
• Earlier onset of action = shorter RT • No need to appeal to complex mental operations of
“slotting”, “phonemic encoding”, etc. • RT difference be explained through embodied processes
Most Important Part…
• A new empirical finding (pre-stimulus onset movement of the mouth) was found by considering ALL of the resources the organism has available, including bodily resources (i.e., articulation)
• Without considering the contribution of bodily processes, the original explanation of the effect was incomplete at best and incorrect at worst.
• We can explain an empirical finding (shorter RT when first letter primed) without appealing to complex cognitive operations and representational abilities!
10/16/2019
13
Research Programs in Embodied Cognition • In your textbook, Shapiro outlines three major
research programs relating to embodied cognition.
Research Programs in Embodied Cognition • Conceptualization
• Concepts and knowledge must relate to embodied processes in order to be represented
• Focus here is on overcoming symbol-grounding problem • In order to “understand” something, it must be related
to perception and action (embodied)
• Symbolic processing is still “allowed” in conceptualization, as long as the symbols are grounded in sensorimotor processes
• Somewhat in competition with standard cognitive science (less emphasis on symbols, processing, and representation)
Research Programs in Embodied Cognition • Replacement
• Attempt to explain all cognitive processes without appealing to symbols or “representation”
• A lot of what we talked about today falls under replacement: outfield problem, resonance, motion parallax as depth determinant
• Emphasis in replacement is on the model of the mind not as a computer or symbolic processor, but a “dynamic system” in continuous motion towards attractor basins
• Can almost be considered a different discipline!
10/16/2019
14
Research Programs in Embodied Cognition • Constitution
• Emphasis on the mind as a “constituent” of cognition… but not the only constituent
• Bodily resources as well as environmental affordances are also considered to be constituents of cognition, and shouldn’t be separated from the mind itself
• Emphasis on embedded and situated cognition – how do people utilize ALL available resources (brain, body, and environment) to perform a specific task
• Not too concerned with shutting down representation and symbolic processing, and can easily “cooperate” with standard cognitive science
