Transcript A dilemma for the extended mind

Physical Symbol System
Hypothesis
A physical symbol system has the necessary
and sufficient means for general intelligent
action. (Newell and Simon, “Computer
Science as Empirical Enquiry...”)
What is a Physical Symbol System?
Five conditions
1. A symbol may be used to designate any
expression (where ‘designate’ means to affect
causally or be affected by);
2. there exist expressions that designate every
process of which the machine is capable;
3. any expression can be created or modified in
arbitrary ways;
4. expressions can be stored without alteration;
5. memory is unbounded.
(One thing these give you is flexible, stored
program computing.)
Modal v. Amodal symbols
A modal symbol is one tied distinctively to a
particular sensory mode or to a motor
program (include bound sensory
representations together).
An amodal symbol is one that is not modal
(these are typically characterized as
arbitrary)
Perceptual Symbol Systems
Human thought consists only of the
manipulation or activation of sensory or
motor states or stored records of them.
On the view of Barsalou et. al, this can
include binding units (conjunctive
neurons), that cause the appropriate
reactivation of the stored records.
Levels of Explanation
We should distinguish between the
cognitive-functional description of the
models and the neural description.
Being modal need not be a claim about
cortex (that, e.g., anything appearing in
sensory or motor cortex is automatically
the physical form of a modal
representation).
“Tied distinctively to...”
What does this mean, then?
It’s partly a claim about processing: that the
way in which symbols are processed is
influenced by the particular modality in
which they appear.
Local v. global representation
Global form assumption: the same symbol
represents a given feature (property) in
connection with all relevant concepts. These are
typically assumed to be amodal (i.e., not tied to
any particular sensory modality).
Local form assumption: distinct symbols are used
to represent the same property in connection
with the various relevant concepts. (These are
frequently assumed to be modal.)
Phenomena that might seem hard
to for the perceptual SS to explain
Categorical inference. Inferences based on the
category something belongs to or relation
between categories :
Sam is a dog. All dogs are animals. All animals can
die. Therefore, dogs can die. Therefore, Sam
can die.
Type-token distinction. We regularly distinguish
between individuals (tokens) and the categories
to which they belong (types).
Abstraction: Forming categories requires ignoring
many properties of individual members.
Productivity and Compositionality
Human language and thought are
generative; an open-ended set of
sentences or thoughts can be generated
from a relatively small number of rules and
unstructured representations.
The content of a thought or statement is a
function of the content of its parts and
combinatorial rules. (Even when it’s not,
the meaning is often a function of
theoretical factors.)
Conjunction detectors
Function as simple markers that reactivate
past records of sensory experience.
Can code for similarities among past
sensory experiences.
This coding explains abstraction and can
explain how the right combinations of
sensory experiences get re-activated.
Reactivation of selected sensory records
can account for productivity and
compositional semantics.
It also accounts for categorical inference
and the type-token distinction.
The particular sensory record is the token;
the type is a collection of aspects
(common features) of particular records.
Status of conjunctive neurons
Read from p. 88.
Are higher-order conjunctive neurons
amodal symbols?
What determines what they activate and
how the results will be read?