Anda di halaman 1dari 1

PROBLEMS!!!!

1) Agency
2)Frame and Binding
3) This account is not as
exhaustive ie there may be a
limit of how much we can
explain of cognitive
architechture due to finite
expressive power

Discrete functions and


systematic breakdown
Uniform competence
within a species

Classical can account


for :specific areas in brain
for specific functions

Mind is Governed by
FORM
Representation of the world
is defined by A priori info.
(this info is the leggo blocks)
The way that the leggo
blocks fit together
determines how the world is
represented. The leggo
proerties have specific
forms
Mix of special purpose and
general purpose domains
ie working memory is a
general purpose

No formal representations
Experience of the outside
world is a series of settled
solutions
No information specific
rules. Generalised ways of
operating

All rules and operations are


specific to the formal
properties of the
informations

info represented by distinct


patterns of activations

4 More complicated
architecture
Evidence for Competence Based
Hypothesis generation
Domains:
belief states. ie current
1) Learnability Problem: Environment is
phenomenal experience is a
impoverished to learn certain things.
hypothesis
Plato perfect world and bring info from
perfect world- A prior knowledge
Input reformatted into a pre2) Systematic constraint and
existed useable form
breakdown: when something fails it does
INDIRECT PERCEPTION
so in a systemataic way ie brocas
We Experience an
aphasia
interpretation of events in
3) Cross cultural consistency: all
the world based on how the
humans can identify certain emotions
leggo blocks have been
4) Illusions:
pieced together
anatomical modularity, uniformity of
functional (mental) architecture across
people, and subtractivity (loss of
Inherent systematicity (due
function, rather than development of a
to form of leggo blocks) and
new replacement function).
Transducer> Input Module
productivity. (endless
(perceptual processes) >
supply of leggo blocks)
Central System (leads to
thought) > output module
An infininte number of
( hold motor routines to
MACRO
thoughts but only of a finite
perform response )> Motor
ARCHITECHTURE: FODOR
type. FINITE EXPRESSIVE
output
MODULARITIY:
POWER
1) Domain Specific
2) competence based
Illusions: Connectionist
Learning= triggering innate
domains
would explain as function of
knowledge (need
3) Informationally
Structures in environment
environmental triggers at a
encapsulated
and how they were
critical period) and
4) indirect representation of
associated .ie experiencesof
hypothesis generation.
the world (computational via
angles and how they are
module)
associated Classical : in
5)attribute non sensory
terms of modules ie
features through
modules are product of
demonstrative inference
evolutionary selection
(can demonstrate why a
particular output occured
due to a partic input) ie
Central system:
depth, gestalts, syntax,
Non-demonstative inference (INDUCTIVE). CAnt predict the
illusions
output from the input
6) Shallow output as not
Belief fixation
privvy to how module works
2 properties:
doman general
takes info form many diff places, brings together and constructs
with its own leggo blocks (mentalese) to form the contents of
consciousness ie the current belief.
1) Isotropic principle: FRAME PROBLEM what is important right
now
2) Quinian principle: anything could be relevant to the truth value
of past beliefs that you have created. eg misinformation
continues to have an effect on future belief states. How do retain
rationlaity when things change

CENTRAL SYSTEM: generates a unique product (belief state) which is


determined by a range of factors, including current beliefs and goals.
Consciousness and meaning both arise out of the operation of this
central system.
Fodor claims that the central system operates along isotropic lines
(any information might be relevant when constructing a particular
belief state) and
Quinean lines (any information might be relevant in assessing the
validity of a particular belief state).
What this means is that the operations of the central system follow no
hard and fast rules and hence we may never find a model to explain

Classical-Plato
evolutionary Psychology
serial processing- 100 step
contraint

Associationist: Aristotle
Behaviourists.

Meaining of current mental


state is defined by what it is
made up of

Competence based domains:


defined by what they do
Modules

Specialist Processess

hard constraints:formal
architectural properties

Bottom up processsing
production systems
ACT-R

Neural Network. Input


(microfeatures) > Energy In
> Weight of connection gets
stronger with each fire >
output layer
Information is the weight of
connections between the
nodes (PDP)

Parallell Distributed processing


Input - Output based on
associative strength

Distributed

Neuro-chemical

We definitely need
information-general
processing, where mental
processes can operate
across a range of
information types in order to
account for the apparent
convergence of information
(from different sources) into
singular belief states (what
you are experiencing now).

Macroarchitechture :Input
- hidden layer - output later
Can have back propagation

Electrical

Meaning of current mental


state is defined by
associations

Event is represented by a
pattern of activation. (as
opposed to leggo blocks)

Knowledge based domains:


deefined by the knowledge
they contain

Infinite expressive power

Schemata
common processes

Learning is change in
connection weights.
Identification is pattern
association

soft contraints: connection


strengths
Top down
connectionist systems
BAckprop, Nettalk

ADVANTAGE :explains
graceful degredation

Problems!!!:
1) One off learning (would
need to change weight of
connections ) and no
inherent constraint on
learning (contradiction
garcias rats)
2) Retroactive interference.
as new thing is learned
are you reducing the
weights of the connections
of other knowlege.
(ie forgetting language
that was last learned
3) If innate information
exists a connectionist
account cant account for
inherent architecture
4) Systematicity : is based
on serendipity

Cognitive support for connectionism


greater parsimony
provides a more general, yet detailed, treatment of learning
and information processing, utilising the same basic
principles
uses soft constraints (connection strengths), as opposed to
hard constraints (formal architectural properties)
allowing for more adaptable cognitions, and an ability to
interpret degraded input, as in the case of categorisation and
learning through experience
Problems:
Humans learn without someone constantly comparing our
output with our input, so is it possible for connectionist
networks to learn independently?

In a connectionist network action is determined by local


resolution (soft constraints) within the network. This is not
going to be the case if the mind is an hypothesis generator.

Anda mungkin juga menyukai