Greg Detre
Tuesday, November 19, 2002
there�s a
really good web page on critiques of gopnik�s paper, including carey + spelke�s
Gopnik:
theories as less impoverished/localised representations of meaning/knowledge
structures
Gopnik +
Carey � both advocates of theory theory
Gopnik �
bi-directional
Kuhn � hard
to objectively measure progress in science, emphasised the social, accidental
and cognitive aspects of science (in the face of a more logicist philosophy of
science approach)
what you see in the data depends on what paradigm you�re working within
influenced by Gestalt and Piaget
Gopnik is
influenced by Kuhn, but revising/attacking Piaget
there�s
this circularity between analogies of theory formation between scientist as
child and child as scientist, when we don�t really know much about theory
formation in either
interesting
but vague � what do they really mean by a �theory� etc.?
Push: model
of the child as a society of scientists?
Gopnik�s
(two distinct) claims:
1.
children�s
knowledge structures are like theories
2.
process
of conceptual development is like the process of theory change in science
she doesn�t want to separate these
what is a
theory?
more than just an �empirical generalisation�
linear relationship between height + weight � is that just an empirical
generalisation?
but they aren�t symmetric � we want to plot height (as independent
variable) along the bottom
but that�s because it�s easier to evaluate height, I�ve seen more things
than I�ve weighed
but it also depends on what problem you�re trying answer
he prefers height along the x, because you think more of weight as a
function of height
but you could also see it in terms of a causal
model, with implicit variables like genetics, childhood nutrition and current
diet � causal models
Gopnik�s now really interested in causal learning in terms of Bayesian
algorithms
considers that they might be innate
a causal model provides explanation
you�ve added the arrowheads, i.e. one-way directedness
no way to distinguish direction of causality based on correlation alone
how much of this discussion arises because they
want theories to be compact/elegant?
or at least a compact core�
tradeoff between complexity + fitting the data
Schultz + Gopnik � experiments to show that
children understand causal logic
mechanisms
of theory change
evidence-driven
relatively orderly, predictable, constrained process
theory formation is an �effective computational procedure�, a �logic of
discovery�
Gopnik doesn�t like neural networks � because
she thinks that learning isn�t incremental, but has discontinuities
isn�t this a misunderstanding about how neural nets work???
she seems to think that the new theory, which had been just applied to
special cases, expands to apply to all � it grows out of the old
example of
theory change � theory of mind
2.5-3 � understand mind in terms of desires and perception, and
perception is veridical
4 year olds � understand mind in terms of desires and beliefs, which may
be false
Deb: is there a similar change when children realise that people may be
masking their desires?
Josh: presumably there could also be a �false desire� test
the mystery
of theory change
with incorrect theory, children ignore or misinterpret counter-evidence
� but what then causes theory change?
they could be keeping track of it in their heads, of course
in science, it could be because you have better technology, or a glut of
counter-evidence � in children, the closest thing you have is improvements in
perceptual/cognitive mechanisms (e.g. short term memory etc.)
why is theory change so much quicker and more regular in development
than science?
because it�s designed
it�s easier
and science is pretty regular (see calculus, evolution etc.)
she argues
against reflection as the origin of theory change
but when you�re building theories, you�re learning methods of theory
revision
in philosophy of science, reflection often comes from thought
experiments
Barbara: how do you know when your theory has
become too complex and needs revision?
Carey wants
to dispute Gopnik�s claim about �theories all the way down� � as deep as you
analyse knowledge that children have, that knowledge has the character of a
theory
does a
theory have to be revisable?
Carey disagrees, saying it doesn�t, because core knowledge is
non-revisable, so if it�s a theory then Gopnik�s definition is too strong
Tom: or is it? what�s wrong with Gopnik�s definition?
Gopnik:
intervention is the key to discovering causal relationships
�did you ever hear of the dog ringing the bell?�
Deb: that doesn�t seem quite right, because his dog used to pull on the
leash to get him to go for a walk, but the dog never pushed on the leash
Garcia effect: rat to learn that a funny taste is associated with
sickness on a single trial, flashing lights with shock, but you can�t get them
to learn other effects so easily � they learn some things more readily than
others � you�re exploiting some other system
Deb: in the
A-not-B error experiments, do they look at eye-tracking?
if you turn out the lights, rather than hiding for them in the wrong
place, very young children reach for the right place, apparently - ???
you can
track up to 4 moving balls, but not 5
Deb: there�s an argument that that�s why there are no verbs in any
language with 5 arguments
Williams
syndrome � don�t have the capacity for conceptual change in many other ways,
but at the age of 4, they pick up the beliefs/desires psychology � Carey: so
it�s as though that specific theory formation mechanism is an embellishment of
the core knowledge system rather than just a general theory
core
knowledge is encapsulated � can�t do abduction
how does
the discussion of children as scientists relate to theory theory and campbell�s
contrast of it (i.e. as some neural substrate for a certain type of
situation-independent tacit knowledge, or something???) with simulation???
can you say
that a theory has to be linguistically expressible???
well, what would that add???
adds abstraction/generalisability, communicability, makes things fixed/precise, formalisable
introspectively feels diff
tracking/Kalman filtering???