Lecture 9: Learning November 5, 1999
Download
Report
Transcript Lecture 9: Learning November 5, 1999
Cognitive Psychology
Part 2: (Behavioral) Learning
I. Learning -- Classical Conditioning
II. Neural Basis of Classical Conditioning
I. Learning
-- Classical Conditioning
1. What is learning?
2. Classical Conditioning
1. What is learning?
Learning is a permanent change in the nervous system
of an organism that changes the way it responds to its
environment, usually as a result of an experience that
the organism went through.
(Note: By learning here we do not mean the acquisition of
knowledge like in school but the acquisition of behavioral and
emotional responses.)
1. Learning is a memory phenomenon.
Without memory, no learning can occur.
2. Learning must be associative in nature.
That is, learning occurs when an association
is formed between two stimuli or between a
stimulus and a response.
Associative Learning
Associative Learning via Bi-directional
Projections from/to Hippocampus
2. Classical Conditioning
A type of learning in which an artificial
stimulus evokes a response that was
originally associated with a natural
stimulus (Ivan Pavlov, 1900?).
(e.g.) A hungry dog
1. Before learning
Food
----->
(natural stimulus)
Salivation
(response)
That is,
Unconditioned Stimulus ------> Unconditioned Response
(UCS)
(UCR)
(‘Unconditioned’ means automatic or without-learning.)
2. During training
Food with bell ----->
Salivation
(natural stimulus with
(response)
artificial stimulus)
UCS + CS ------>
UCR
Note that the Salivation (UCR) is evoked by food
(UCS), not by the bell (CS).
3. After learning
Bell
----->
Salivation
(artificial stimulus only)
(response)
CS
CR (not UCR)
------>
An association that has been formed between the CS
(bell) and UCR (salivation) during training evokes the
same salivation response (now termed as CR), without
the participation of the UCS (food) .
What association and how?
Everyday Examples of Classical Conditioning
1. Advertising
During training
Man in cowboy hat (UCS)
and cigarette (CS) -------> Favorable feeling (UCR)
After training
Cigarette (CS) in supermarket ------> ???
(also, woman in bikini + beer; wealthy looking
business man + car, etc.)
2. Emotion
Robbery at gun point (UCS)
in dark alley (CS) ----> Frightening experience (UCR)
Dark alley (CS) ------> Get frightened (CR)
Also,
Demanding ill-tempered supervisor with red mustache
----> bad experience and eventually got fired
3. Physiological response
Harmless Turkish food (CS) contaminated with a
poisonous chemical of unknown origin (UCS)
----> Severe stomach ache (UCR)
Turkish food (CS) ------> Stomach ache (CR)
Un-awared self-curing capacity (or simply disease gone
through its course of action) + tooth paste
---> disease cured (“placebo effects”)
Instead of tooth paste, how about simple meditation,
bible reading or even reciting of incomprehensible code
words?
II. Neural Basis of
Classical Conditioning
All that is needed to explain the
classical conditioning phenomenon
is the associative Hebbian learning
rule.
1. Before learning
Wucs >> 0; Wcs = 0,
UCS alone evokes the salivation (UCR)
2. During training
UCS
(present)
CS
(present)
R
(present)
Hebbian learning then implies that
DWcs > 0
on every training trial.
3. After learning
At the end of training,
Wcs >> 0 (sufficiently large)
which means that
CS alone can now evoke the salivation (CR)!
Neurophysiology of Conditioning
More sophisticated models of conditioning
Discussion of the Paper “Is the Brain a Digital Computer” by J. Searl
Cognitivism as DCTM (Digital Computational Theory of Mind)
(excepts from R M Harnish (2002, chap. 8)
Point #1 (Cognition as Symbol Manipulation):
Cognition, in essence, is digital computation that involves manipulating
(creating, transforming and deleting) symbols which we call mental
representations of the world around us.
Point #2 (Symbol Manipulation Defines Computation):
According to Turing, computation is defined syntactically in terms of symbol
manipulation.
Point #3 (Brain as a UTM)
Cognition, or thinking for that matter, really is a species of digital computation,
then the Church-Turing thesis tells us that these can be carried out by Turing
machines or just one Universal Turing machine. In other words, the brain is
somewhat equivalent to a Universal Turing machine, say a von Neumann
machine or a production system or the like.
Point #4 (Cognitivist Scientific Research Paradigm):
So, the reasoning goes, therefore, the job of cognitive science, neuroscience,
and artificial intelligence is to discover the actual algorithms of the mind.
Q1: Is this a Strong or Weak AI view?
Q2: Do you think Searl endorses the DCTM?
Q3: What’s Searl answer to the question “Is the brain a digital
computer?” What are the arguments he puts forward to defend his
position?
- Chinese Room Argument (NP-hard problem??)
- Syntax is not the same as semantics;
Semantics is not intrinsic to syntax
- Homunculus fallacy
Q4???