Transcript Operant conditioning

Operant conditioning
Operant conditioning
In classical conditioning, the presence of one stimulus
(e.g. meat powder) is conditional on the presence of
another stimulus (e.g., a bell)
What else can an animal learn, besides the relationship
of two stimuli?
Operant conditioning
It is also possible for the animal to generate a response
and for that response to have consequences:
Act cute, you get pet
Poop on the rug,
you get scolded
Note that the thing to be learned is not a
UR. Animal emits a response (pooping,
acting cute), and it is rewarded or
punished.
Edward Thorndike
Thorndike’s Law of Effect
“If a response in the presence of a
stimulus is followed by a satisfying
event, the association between the
stimulus and the response is
strengthened. If the response is
followed by an annoying event, the
association is weakened.”
Today we’ll cover:
• Basics of operant conditioning
• What makes operant conditioning
effective.
• The problem of definition in o.c.
(not just that “animals seek
rewards”).
Thorndike’s method was limited
because each trial took so long.
A stripped-down environment
Free operant curve, from a
cumulative recorder
Steep slope=many responses
Shallow slope=few responses
What would the curve look like if 20
bar presses food?
To really teach the animal you would
shape it’s behavior. . .
Fixed ratio
Consistent ratio of number of responses &
number of reinforcers
Steady response
Easy to extinguish
Example: factory
Piece work
Variable ratio
Set ratio of number of responses &
number of reinforcers, but can vary locally
Example: slot machine
Rapid response
Hard to extinguish
Fixed interval
First response after a specific amount of
time since the last reinforcement
Example: studying
for exams
Little response until
just before
reinforcement:
then rapid response
Fairly easy to
extinguish
Variable interval
First response after a some amount of time since the
last reinforcement: amount of time can vary, locally
Example: checking
email
Steady response
Hard to extinguish
Contingencies don’t just add good stuff. . .
Result
e.g., food
Add to
environment
Increase
probability
of behavior
Positive
Reinforcement
Punishment
Negative
Reinforcement
Negative
Punishment
(Extinction)
Action
Take away
from
environment
e.g., escape
e.g., spanking
Decrease
probability
of behavior
e.g., being grounded
Complex contingencies:
Would this work?
Bar press
reinforced,
but ONLY
when red
light is on.
YES! This is
called differential
reinforcement
How does differential
reinforcement apply here?
Reinforcer = food.
Response = hovering
Differential signal = looking up.
What’s happening, and what should the birds do?
What’s happening= differential sign has changed
What should the birds do = stop responding
Moments later, birds are leaving
Operant conditioning--what
makes it effective?
•
•
•
•
•
Schedule of reinforcement
Temporal contingency
Belongingness
Quality, quantity of reinforcer
What else the animal might do
T-maze: temporal contingency
Condition 1: immediate reward (.5 sec)
!
Condition 2: delayed reward (5 sec)
!
Effectiveness--temporal contingency
Learning strength (arbitrary
units)
The delay between the animal’s act that you
are reinforcing, and the reinforcer.
35
30
25
20
Grice (1948)
Wolfe (1934)
15
10
5
0
0
5
10 15 30 60 150
Delay (seconds)
WHY does learning drop off with
delay??
Condition 2: delayed reward (5 sec)
!
Operant conditioning--what
makes it effective?
•
•
•
•
•
Schedule of reinforcement
Temporal contingency
Belongingness
Quality, quantity of reinforcer
What else the animal might do
Belongingness
• Thorndike tried to condition his cat to yawn or
scratch to escape box--he proposed belongingness
Instinctive drift
• A concept related to belongingness:
instinctive drift (Breland & Breland.)
•Motivational state can also influence; a
hungry animal does more food-seeking
behaviors. . .
Digging
Digging, scratching, rearing
Quality/quantity of reinforcer
Works as you would expect.
What else might the animal do?
It’s not as simple as “the animal
Maximizes good things, minimizes”
bad things.
Even humans don’t do this, if the situation gets
moderately complex.
Example
Variable ratio
Variable interval
What’s the optimal strategy?
Variable ratio
Variable interval
Optimal is to hit VR almost exclusively
and occasionally hit the VI. Instead, they
respond to equalize ratios of work/reward
The problem of definition
What is a reinforcer?
The problem of definition
Thorndike called a reinforcer something “that
brings about a satisfying state of affairs.”
How do we know when animal is satisfied?
Presumably, when the animal will work to
achieve this state of satisfaction.
But that’s circular
What’s a
reinforcer?
What’s
pleasurable?
Something
pleasurable.
Something that
increases
behavior, that
animal will work
to get
What will the
animal work for
(e.g., peck)?
Another definition: physiological
homeostasis
Animal seeks to lessen thirst, hunger, etc.
Definition of reinforcement is based on biological
drives.
Learning = a “stamping in” of the work that needs to
be done to reduce hunger.
E.g, “I must not only consume and chew to get
nourishment. I also must press the bar, then consume,
then chew.
Problems
Too many drives were proposed.
Animals (and people) do things that seem
more likely to raise drives, not lower
them
Reinforcement as behavioral
regulation
Premack principle: Given two
responses arranged in an operant
conditioning procedure, the more
probable response will reinforce the
less likely behavior.
Which do you want to do: play
pinball or eat candy?
Must eat candy
to play pinball
These kids treat candy eating
as work: do it to get to play
pinball.
Must eat candy
to play pinball
These kids eat candy but
don’t care that they have
earned pinball time.
Behavioral homeostasis & bliss
point—a clever, not-quite-right idea
Minutes drinking
35
30
25
20
Restricted
drinking
15
10
5
0
0
15
30
45
60
Minutes running
Normally, animal likes to be at gray spot (15 minutes of
each--now it can’t be at gray spot. What will it do?
Minutes drinking
35
30
25
20
Restricted
drinking
15
10
5
0
0
15
30
45
60
Minutes running
IN THEORY you should be able to predict what animal will
do--it will select spot on blue line that is as close as
possible to it’s “bliss point”. IN REALITY this prediction
sometimes works, sometimes doesn’t.
Reinforcement--final word
In the end, we still don’t have a good
definition of the concept. Premack
Principle is as close as we get.
Nevertheless, the concept of
reinforcement seems useful.
Applications
•
•
•
•
Animal training
Biofeedback
Education
Token Economies
Biofeedback
Operant conditioning of the autonomic
nervous system.
For years, not explored because no one
thought it could possibly work.
Apply operant conditioning principles
to education
1. Make sure student doesn’t make mistakes; guide
behavior.
2. Review frequently.
Little enthusiasm. Teachers didn’t like it
for their own reasons. Students were bored.
Token economies
Used in some mental health institutions,
and some classrooms.
Mrs. Ahlersmeyer’s 3rd grade class,
Lafayette Elementary, Lafayette, IN
• Students earn a “salary” (marbles).
• Outstanding work or behavior earns bonuses.
• Students allowed 5 sick days per quarter, after
that, they are docked pay.
• Students charged rent for their use of desk, and for
any school property lost or damaged.
• Students docked pay for inappropriate behavior.
Use is controversial because it seems
“dehumanizing” (mental patients) or
because it seems that you’re
“paying” students for behavior that
they should want to do.
Applications
•
•
•
•
Animal training
Biofeedback
Education
Token Economies