LINKAGES Networks of Learning

in review
190
Chapter 5
Learning
REINFORCEMENT AND PUNISHMENT
Concept
Description
Example or Comment
Positive
reinforcement
Increasing the frequency of a behavior by following
it with the presentation of a positive reinforcer—
a pleasant, positive stimulus or experience
You say “Good job!” after someone works
hard to perform a task.
Negative
reinforcement
Increasing the frequency of a behavior by following
it with the removal of an unpleasant stimulus or
experience
You learn to use the “mute” button on the TV
remote control to remove the sound of an
obnoxious commercial.
Escape conditioning
Learning to make a response that removes an
unpleasant stimulus
A little boy learns that crying will cut short
the time that he must stay in his room.
Avoidance
conditioning
Learning to make a response that avoids
an unpleasant stimulus
You slow your car to the speed limit when
you spot a police car, thus avoiding being
stopped and reducing the fear of a fine;
very resistant to extinction.
Punishment
Decreasing the frequency of a behavior by either
presenting an unpleasant stimulus (punishment 1)
or removing a pleasant one (punishment 2,
or penalty)
You swat the dog after it steals food from
the table, or you take a favorite toy away
from a child who misbehaves. A number of
cautions should be kept in mind before
using punishment.
?
1. Taking an aspirin can relieve headache pain, so people learn to do so through the process of
reinforcement.
2. The “walk” sign that tells people it is safe to cross the street is an example of a
stimulus.
3. Response rates tend to be higher under
schedules of reinforcement than under
schedules.
Online Study Center
Improve Your Grade
Tutorial: Reinforcement
and Punishment
more productive lives (e.g., Alberto, Troutman, & Feagin, 2002; Pear & Martin, 2002).
These programs include establishing goal behaviors, choosing reinforcers and punishers,
and developing a systematic plan for applying them to achieve desired changes. Many
self-help books also incorporate principles of positive reinforcement, recommending
self-reward following each small victory in people’s efforts to lose weight, stop smoking,
avoid procrastination, or reach other goals (e.g., Grant & Kim, 2002; Rachlin, 2000).
When people cannot do anything to alter the consequences of a behavior, discriminative stimuli may hold the key to changing that behavior. For example, people often
find it easier to quit smoking if, at first, they stay away from bars and other places where
there are discriminative stimuli for smoking. Old cues can trigger old behavior, so they
should avoid the old cues until new behavior can be established. Stimulus control can
also help alleviate insomnia. Insomniacs tend to use their beds for activities such as
watching television, writing letters, reading magazines, worrying, and so on. Soon the
bedroom becomes a discriminative stimulus for so many activities that relaxation and
sleep become less and less likely. Stimulus control therapy encourages insomniacs to use
their beds only for sleeping, and perhaps sex, making it more likely that they will sleep
better when in bed (Edinger et al., 2001).
A
LINKAGES
ssociations between conditioned
stimuli and reflexes or between
Networks of Learning
responses and their consequences
play an important role in learning, but how
are they actually stored in the brain? No one yet knows for sure, but associative network models provide a good way of thinking about the process. As suggested in the
191
Instrumental and Operant Conditioning: Learning the Consequences of Behavior
FIGURE 5 .1 3
An Associative Network
Tree
Here is an example of a network of associations to the word “dog.” Network theorists suggest that the connections shown
here represent patterns of connections
among nerve cells in the brain.
Cat
Snake
Seal
Friend
Pet
Dentist
Bark
Companion
O
Door
Daisy
D
Pain
G
Danger
Fire
Danger
Bite
Insect
Protection
Guard
LINKAGES
How are learned associations
stored in memory? (a link to
memory)
Lock
Police
chapter on memory, the associations we form among stimuli and events are represented in complex networks of connections among brain cells, or neurons. Consider
the word dog. As shown in Figure 5.13, each person’s experience builds many associations to this word, and the strength of each association will reflect the frequency
with which dog has been mentally linked to the other objects, events, and ideas in
that person’s life.
Using what they know about the laws of learning and about the way neurons communicate and alter their connections, psychologists have developed computer models of how
these associations are established (Messinger et al., 2001). An important feature of these
parallel distributed processing models is the idea of distributed memory or distributed
knowledge. These models suggest, for example, that your knowledge of “dog” does not lie
in a single spot, or node, in your brain. Instead, that knowledge is distributed throughout
the network of associations that connect the letters D, O, and G, along with other dogrelated experiences. In addition, as shown in Figure 5.13, each of the interconnected nodes
that make up your knowledge of “dog” is connected to many other nodes as well. So the
letter D will be connected to “Daisy,” “Danger,” and many other concepts.
Neural network models of learning focus on how these connections develop through
experience (Hanson & Burr, 1990). For example, suppose you are learning a new word
in a foreign language. Each time you read the word and associate it with its English
equivalent, you strengthen the neural connections between the sight of the letters forming that word and all of the nodes activated when its English equivalent is brought to
mind. Neural network, or connectionist, models of learning predict how much the
strength of each linkage grows (in terms of the likelihood of neural communication
between the two connected nodes) each time the two words are experienced together.
The details of various theories about how these connections grow are very complex,
but a theme common to many of them is that the weaker the connection between two
items, the greater the increase in connection strength when they are experienced
together. So in a classical conditioning experiment, the connections between the nodes
that characterize the conditioned stimulus and those that characterize the unconditioned stimulus will show the greatest increase in strength during the first few learning trials. Notice that this prediction nicely matches the typical learning curve shown
in Figure 5.3 (Rescorla & Wagner, 1972).
Neural network models have yet to fully explain the learning of complex tasks, nor
can they easily account for how people adapt when the “rules of the game” are suddenly changed and old habits must be unlearned and replaced. Nevertheless, a better
understanding of what we mean by associations may very well lie in future research on
neural network models (Anthony & Bartlett, 1999; Goldblum, 2001).