Transcript Operant versus classical conditioning: Law of Effect

Operant Conditioning
The Law of Effect
• Thorndike (1911): Animal Intelligence
– Experimented with cats in a puzzle box
• Put cats in the box
• Cats had to figure out how to pull/push/move lever to get
out; when out got reward
• The cats got faster and faster with each trial
• Law of Effect emerged from this research:
– When a response is followed by a satisfying state
of affairs, that response will increase in frequency.
E.L. Thorndike
1874-1949
Skinner’s version of
Law of Effect
• Had two problems with Thorndike’s law:
– Defining “satisfying state of affairs”
– Defining “increase” in behavior
• Rewrote the law to be more specific:
– Used words reinforcer and punisher
– Idea of reinforcer is strengthening of relation between a R and Sr
• Now defined reinforcement and punishment:
– A reinforcer is any stimulus which increases the probability of a response
when delivered contingently
– A punisher is any stimulus which decreases the probability of a response
when delivered contingently
– Also noted could deliver reinforcers and punishers in TWO ways:
• Add something: positive
• Take away something: negative
Burris Fredric Skinner
Skinner box:
Pigeon pecks or rat bar presses to receive reinforcers
Reinforcers vs. Punishers
Positive vs. Negative
•
•
•
•
Reinforcer = rate of response INCREASES
Punisher = rate of response DECREASES
Positive: something is ADDED to environment
Negative: something is TAKEN AWAY from
environment
• Can make a 4x4 contingency table
Reinforcement
Punishment
Positive
Add
Stimulus
Positive Reinforcement
make bed-->10cent
(Positive) Punishment
hit sister->spanked
Negative
Negative Reinforcement
Negative Punishment
Remove
make bed-> Mom stops
hit sister->lose TV
Stimulus
nagging
Parameters or Characteristics of
Operant Behavior
• Strength of the response:
– With each pairing of the R and Sr/P, the
response-contingency is strengthened
– The learning curve is
• Monotonically ascending
• Has an asymptote
• There is a maximum amount of responding the
organism can make
Parameters or Characteristics of
Operant Behavior
• Extinction of the response:
– Remove the R Sr or RP contingency
– Now the R  0
• Different characteristics than with classical
conditioning:
– Animal increases behavior immediately after the
extinction begins: TRANSIENT INCREASE
– Animal shows extinction-induced aggression!
– Why?
More parameters:
• Generalization can occur:
– Operant response may occur in situations similar to the one in
which originally trained
– Can learn to behavior in many similar settings
• Discrimination can occur
– Operant response can be trained to very specific stimuli
– Only exhibit response under specific situations
• Can use a cue to teach animal:
– S+ or SD : contingency in place
– S- or S : contingency not in place
– Thus: SD: RSr
Several ways to use
Operant Conditioning
• Discrete Trial Procedures:
– Has a set beginning and end
– a trial
– The experimenter controls the rate of behavior and
reinforcement
• Free operant procedures
– Session has a start and end
– Organism can make as many responses as ‘wants” in session
– Organism controls how many reinforcers earned (according
to programmed schedule)
– Organism controls rate of responding: R’s/min
Must “train” or “teach” an
operant response
• in lab setting: magazine training
–
–
–
–
Train animal to come to feeder
Really is classical conditioning
Click of feeder predicts food
No response (other than approach) required
• Shaping
–
–
–
–
Using successive approximations of the final response
Break up a response into its components or pieces
E.g., tying a shoe: how many steps?
Train so that put the “steps” together until have the fluid final response
• For example: Clicker Training!
Why Shaping?
•
Is fast and efficient way to develop new behavior
•
Maintains the learner’s excitement and willingness to learn
•
Produces behaviors that are accurately remembered, unlike coerced or lured responses
•
Allows on to train individuals and behaviors not easily trained in more traditional ways
•
Creates empathy in animals and allows one to read and understand the animal’s emotions
•
Changes the organism being shaped from a passive recipient of information and guidance to an
active learner and member of the learning team
Is fun for both the animal and the trainer!
•
How to Shape Behavior
• Start with Capturing a behavior or component of the behavior you
want to shape
–
–
–
–
Allows you to selectively reinforce some of the repetitions, but not others
Can select bigger, better responses
Ignore weaker responses: these go unclicked
Leads to a high rate of reinforcement
• High rate of reinforcement CRITICAL for shaping!!!!
• Raising the Criterion:
– Raise the criterion or rule for getting a C/T
– Build the response in a series of small steps
– Think of it as going up a staircase towards your goal.
The Shaping Staircase
Each step on the staircase is an increment or step towards the final be
The Shaping Process
• Each step in the shaping process may have
several requirements:
– Every requirement within a step is called a criterion
(pl. = criteria)
– Criteria can be very general or very detailed
• For the high five:
– The lifting of the paw is a more general criterion.
– The touching of the paw to your hand is more
specific
Example of Shaping:
Clicker Training
• Popular term for training/teaching method of
operant conditioning
– Can be used with any living organism
• Gold fish
• Dogs
• Humans!
• Very simple process:
– S+ RSrcSr
– Cue response markerreinforcement
Clicker training or Tag Teach
• System of training/teaching that uses
positive reinforcement in combination with
an event marker
• The event marker (click) “marks” the
response as correct
Why should you use a clicker?
• Very powerful teaching tool
• According to Karen Pryor, clicker training
–
–
–
–
–
–
Accelerates learning
Strengthens the human-animal bond
Produces long term recall
Produces creativity and initiative
Forgives your mistakes
Generates enthusiastic learners
Examples of learning vs.
environmental manipulation
• Want to keep dog out of kitchen:
– Put up a gate: dog can’t get in, so behavior decreases
– Does not alter the contingency of going into the kitchen
– The dog has learned nothing
• Want you to sit in a chair
– I poke you behind the knees and you fall into the chair
– You increased “chair sitting” but didn’t learn chair sitting!
– Your behavior is not predictable when presented with the
chair
– or worse yet, you are now afraid of the chair and avoid it!
Which consequence should we use?
• Punish the behavior?
– Decreases the probability of the behavior
– Can result in unstable responding, particularly
with negative reinforcement
– Can result in learned helplessness, avoidance
and aggression!
– Often are ethical limitations
Which consequence should we use?
• Ignore the behavior?
– Decreases the probability of the behavior
– Process of extinction
– Two problems:
• Extinction burst
• Extinction-induced aggression
Which consequence should we use?
• Positively Reinforce a behavior:
– Increases the probability of the behavior
– Can reinforce the opposite of the response you are
trying to decrease!
– Creates a “fun” learning environment
– Data suggest that organisms trained with positive
reinforcement WANT to work!
Which consequence should we use?
• But wait: won’t positive reinforcement make greedy
organisms?
– Initially, we use continuous reinforcement
– Gradually we thin out the rate of reinforcement using
partial schedules of reinforcement
– More and more responding or chains of behavior
required to get a reward
• When you were kindergarten, you needed lots of
reinforcers every day
• Now in college you can work all semester for that
final reinforcer of an “A”.
What skills are necessary to become
a good shaper/trainer?
•
Must be an excellent observer of behavior
–
•
Must be precise with your clicker
–
–
•
When learning a new response, the animal needs lots of feedback
Reinforcement variety improves the learning process
Reinforcers must be of value to the learner (what THEY like, not what YOU like).
Barney stickers or Beer?
Must use the clicker as a conditioned reinforcer:
–
–
–
•
Must be quick and “catch” and “mark” that response
May introduce a “keep going” signal, too!
Must be generous with reinforcement
–
–
–
–
•
Must be able to identify the response or component of the response
The clicker derives value from it being tightly paired with the primary reinforcer
Use as a bridge or a “yes, keep going” signal
We will call the clicker an event marker
Must be consistent!
–
–
The animal is learning the rule, so the rule must be consistent
Only when the response is solid will you move to partial reinforcement