Transcript Slide Set 3a

Information Processing Theories of
Development
*Jesse Wilkinson*
A general note…
 Connectivism and Evolutionary Cognitive
Development are two approaches to development
that rely on analogies to biological phenomena
 Connectivism: brain structure
 Evolutionary Cognitive Development: evolution/natural
selection
 [We will examine some parallels between these
approaches and their biological counterparts
throughout…]
Connectivism
 An approach to explain mental phenomena via
of thinking, which resemble the brain’s basic
structure
 This is appealing and powerful because “building a model”
allows us to test theories
Connectivist Model in Action
*Spreading Activation*
Organization of Connectivist Models
[and how it corresponds to the brain]
 Simple information processing units
[neurons]
 Units organized in layers
[as are neurons in the brain- example: cerebral cortex
contains 6 layers]
 Input units: contain information about the
initial representation [sensory neurons]
 Hidden: combine units of evaluation criteria
[all other neurons]
 Output: determine response [motor neurons]
 Units are interconnected
[synapses]
Propagation of Information via Connectivist
Models
[brain considerations]
 Output depends on two things:
 (1) Combined activation received from each interconnecting unit (parameters for
processing are set by researcher)
 (2) Connection strength: how strongly or weakly this connection should be tied to the
output (based on the model’s past experience)
 In order for output to be propagated, a threshold must be met, which is predetermined by researcher.
 Exact value of the output activation can fall anywhere within the range of 0 to 1, 1
being strongest.
[Note: this not totally analogous to action potentials (neural output signals) because not “all or none”]
 Many units are activated simultaneously (or in parallel)
[as is the case with neural processing]
 Information is distributed throughout the units (i.e., no single location corresponds
to a particular piece of knowledge).
MacWhinney model
 Tested a system’s ability to chose one of six correct articles
(“the”) in German which change depending on the
corresponding noun and context
 Input layer: 35 units
 Features of the noun: aspects of its sound, meaning, and
context
 Hidden layer: combinations of the 35 input units
 Output layer: the 6 articles (der, die, das, etc…)
Training explained via MacWhinney
102
common
German
nouns
presented
Model
responded
Correct
answer
presented
Model
adjusted
connection
strengths to
optimize
future
accuracy
> 90% correct responses, but is it
really human-like?
Maybe….
 Over-uses articles that accompany feminine nouns (more
common), just like German-speaking children.
 Combinations hardest to learn for kids were hardest to learn
for model.
Connectivist Models & Learning
*Learning occurs through comparing correct responses
with incorrect responses and adjusting the strength of
associated connections until eventually the model
captures complex patterns of multiple, interacting cues.
*Another example: Deep Blue
 Generalization of the system’s knowledge is based on
how similar a new situation is to ones the system has
encountered previously.
Connectivist models have successfully
mimicked many developmental
phenomena






Object Permanence
Understanding time-speed-distance problems
Early reading acquisition
Second language learning (and the critical period)
Category learning
Grammar
Some Clinical Applications
 Influenced the field’s general conceptualization of developmental
disorders by distinguishing them from adult brain damage
 Dyslexia has been simulated in a number of ways:





reducing the number of hidden units
slower rate of connection weight change
Constraining the size of weights in learning
Eliminating connections
Simply exposing the system to less training
 Features of Autism have been simulated:
 Decreases in the number of hidden units (failure to learn in complex
domains)
 Or Increases! (fast initial learning that later regresses)
Limitations
 Generally such models simulate learning, as opposed to
development
 Highly task-specific
 Over-simplified/Reductionist (Biology is NOT math!)
 Structure not really analogous to the brain
 No mention of chemical activity (neurotransmitters)
 all cognition can’t be explained by neural activity
 The “Behavior” of models is not really human-like:
 Require more exposures than humans
 Do not show insight
 Do not learn symbolic rules (like mathematical formulas)
Cognitive Evolutionary Theory
Correct responses
 This is an approach to explain cognitive processes following basic ideas of
Darwinian evolution.
 In studying evolution and development, the fundamental question is the
same: How does change occur?
 Siegler’s Answer: There is competition is among ideas/strategies and this
leads to adaptive outcomes over time.
Age
Siegler’s overlapping waves model of
cognitive development
 At any one time children have
many (competing) ways of
thinking about most topics
 With experience, some become
more/less frequent
 NOT a series of distinct steps
 With time, more advanced
strategies prevail
Strategy Selection
* Experience is key…
* Provides not only answers to the problems, but also information about
speed and accuracy of strategy utilized.
 This information “feeds back” to provide increasingly detailed
knowledge for future strategy selections.
* With experience…
* Children tend to use each strategy most often on problems where it
works especially well compared with alternative approaches [strategies
“find their niches”]
* The more effective something has been in the past, the more often it will
be chosen in the future [“survival of the fittest”]
 ULTIMATE GOAL…
 Retrieval (i.e., to get to a point where using a strategy is
unnecessary). It is faster and just as accurate!
How do children learn new strategies?
 Any number of possibilities:
 Via teaching
 Via imitating others
 Via Spontaneous Strategy Discovery
Siegler’s Approach
 Because he rejects the stage-approach, Siegler also feels that
typical methods used to study development are inadequate.
 Siegler advocates that to observe cognitive change, we need data
collected at brief periods, repeatedly from the same individuals.
 This yields richer, more meaningful data.
What are some examples of strategies kids use to
help them with addition problems?
 Counting from one on their fingers
 Putting appropriate # of fingers up and then counting
them to arrive at an answer
 Using memory
 Guessing!
 “Counting on” (by choosing the larger number first
and counting up with the smaller number)
Siegler’s Study
 Subjects were four- and five-year-olds (N = 8) with some skill in
adding numbers, but did not yet know or utilize “counting on”
strategy
 11 week practice period where the children were presented
with addition problems, three times per week
 Initial problems used only numbers 1-5 (Case), but challenge
problems (e.g., 21+3) were added to create problems where
“counting on” was necessary.
What strategies did the kids use?
 Variability was noteworthy:
 Every child used at least 5
strategies [competition exists!]
 Reliance on particular
strategies varied greatly by
individual
 This was not fully explained by
more knowledgeable kids
using more advanced
strategies (the kid with highest
ranking correct ranked 4th on
use of retrieval).
(Note: “Counting on” is also known as the “min method”)
7/8 children eventually discovered
“counting on”
 Siegler observed what led up to strategy discovery:
 The only distinguishing characteristic prior to discovery was a long
solution time compared to child’s mean solution time.
 Discovery also accompanied by indicators of cognitive confusion:
false starts, pauses, and odd statements
Possible mechanisms
 Discoveries occur in response to impasses or failures?
 Nope! Most discoveries were made on problems the kid had previously solved without difficulty;
preceding problems were not unusually difficult
 Transition strategies lead to discovery (which contain some but not
all of the elements of the ultimate strategy)?
 Short-cut sum strategy
 4+2 = “1, 2, 3, 4…5, 6”; as opposed to 4+2 = “1, 2, 3, 4… 1, 2,…1, 2, 3, 4, 5, 6”
 Yes indeed- this strategy emerged in close proximity BEFORE discovering “counting on” in all 7
Contrasting examples…
Brittany & Whitney
Once discovered, there were large
individual differences in level of
insight, awareness, and affective
reactions…
“Eureka!”
Siegler & Crowley, American Psychologist, 1991
“Huh?”
Generalization of the Strategy
 Some lacked insight, but even children who clearly
articulated the strategy did not immediately apply it.
 The two children who ended up using it most only “counted
on” for 7/84 and 2/49 or their trials following their discovery.
 Children’s emerging knowledge may be implicit and not
accessible to verbal report.
 Experience may be necessary both before children fully utilize
strategies and are able to articulate them.
 This supports “Wave theory”: change does not occur
immediately following discovery.
Testing the transition mechanism and
generalization
 Siegler used a computer model that:
 Analyzes the sequence of operations involved in executing strategies
 Identifies potential improvements (e.g., redundancies)
 Generates new strategies by combining old ones
It worked!
 The computer model discovered the “counting on” strategy!
 The model’s behavior was similar to children:
 Sometimes discovered strategy following incorrect performance, sometimes correct
 Generalized new strategies in a similar fashion
Siegler’s findings have been replicated in explaining the
acquisition of other developmental skills:






Time-telling
Reading
Spelling
Tool use
Problem-solving
Memory tasks
Limitations
 Limited applicability:
 Theory is most applicable to domains in which children use
clearly defined strategies
 Says little about how the social world influences
cognitive development
THANK YOU!!