Pavlovian Conditioning
|
Reinforcement (behavior increases) |
Punishment (behavior decreases) |
|
Positive (something added) |
Positive Reinforcement: Something added increases behavior |
Positive Punishment Something added decreases behavior |
Negative (something removed) |
Negative Reinforcement Something removed increases behavior |
Negative Punishment Something removed decreases behavior |
Remember that these definitions are based on their actual effect on the
behavior in question: they must reduce or strengthen the behavior to be
considered a consequence and be defined as a punishment or
reinforcement. Pleasures meant as rewards but that do not strengthen a behavior are indulgences, not reinforcement; aversives meant as a behavior weakener but which do not weaken a behavior are abuse, not punishment.
This
is possibly the easiest, most effective consequence for a trainer to
control (and easy to understand, too!). Positive reinforcement means
starting or adding Something Good, something the animal likes or
enjoys. Because the animal wants to gain that Good Thing again, it will
repeat the behavior that seems to cause that consequence.
The
dolphin gets a fish for doing a trick. The worker gets a paycheck for
working. The dog gets a piece of liver for returning when called. The
cat gets comfort for sleeping on the bed. The wolf gets a meal for
hunting the deer. The child gets dessert for eating her vegetables. The
dog gets attention from his people when he barks. The elephant seal
gets a chance to mate for fighting off rivals. The child gets ice cream
for begging incessantly. The toddler gets picked up and comforted for
screaming. The dog gets to play in the park for pulling her owner
there. The snacker gets a candy bar for putting money in the machine.
A
primary positive reinforcer is something that the animal does not have
to learn to like. It comes naturally, no experience necessary. Primary
R+s usually include food, water, often include sex (the chance to
mate), the chance to engage in instinctive behaviors, and for social
animals, the chance to interact with others.
A secondary
positive reinforcer is something that the animal has to learn to like.
The learning can be accomplished through Classical Conditioning or
through some other method. A paycheck is a secondary reinforcer - just
try writing a check to reward a young child for potty training!
Animal trainers will often create a special secondary reinforcer they
call a bridge. A bridge is a stimulus that has been associated with a
primary reinforcer through classical conditioning. This process creates
a conditioned positive reinforcer, often called a conditioned
reinforcer or CR for short. Animals that have learned a bridge react to
it almost as they would to the reward that follows (animals that have
learned what clicker training is all about may sometimes prefer the CR
that tells them they got it right over the actual "reward").
A schedule of reinforcement determines how often a behavior is going to result in a reward. There are five kinds: fixed interval, variable interval, fixed ratio, variable ratio, and random.
A fixed interval means that a reward will occur after a fixed amount of time. For example, every five minutes. Paychecks work on this schedule - every two weeks I got one.
A variable interval schedule means that reinforcers will be distributed after a varying amount of time. Sometimes it will be five minutes, sometimes three, sometimes seven, sometimes one. My e-mail account works on this system - at varying intervals I get new mail (for me, email is generally a Good Thing!).
A fixed ratio means that if a behavior is performed X number of times, there will be one reinforcement on the Xth performance. For a fixed ratio of 1:3, every third behavior will be rewarded. This type of ratio tends to lead to lousy performance with some animals and people, since they know that the first two performances will not be rewarded, and the third one will be no matter what. Some assembly-line production systems work on this schedule - the worker gets paid for every 10 widgets she makes. A fixed ratio of 1:1 means that every correct performance of a behavior will be rewarded.
A variable ratio schedule means that reinforcers are distributed based on the average number of correct behaviors. A variable ratio of 1:3 means that on average, one out of every three behaviors will be rewarded. It might be the first. It might be the third. It might even be the fourth, as long as it averages out to one in three This is often referred to as a variable schedule of reinforcement or VSR (in other words, it's often assumed that when someone writes "VSR" they are referring to a variable ratio schedule of reinforcement).
With a random schedule, there is no correlation between the animal's behavior and the consequence. This is how Fate works.
If reinforcement fails to occur after a behavior that has been reinforced in the past, the behavior might extinguish. This process is called extinction. A variable ratio schedule of reinforcement makes the behavior less vulnerable to extinction. If you're not expecting to gain a reward every time you accomplish a behavior, you are not likely to stop the first few times your action fails to generate the desired consequence. This is the principle that slot machines are based on. "OK, I didn't win this time, but next time I'm almost sure to win!"
When a behavior that has been strongly reinforced in the past no longer gains a reinforcement, you might experience what's call an extinction burst. This is when the animal performs the behavior over and over again, in a burst of activity. Extinction bursts are something for trainers to watch out for!
Recently Bob Bailey has
cautioned against needlessly using variable schedules. Most useful
behaviors, he points out, will get some sort of reinforcement every
time. You might not always click and treat your dog for sitting on cue,
but you will always reward it with some recognition and praise ("Good
dog!"). If there is some circumstances where you will be unable to
deliver any reinforcement (during a long sequence of behaviors, or when
the animal is out of contact), then you will need to build a buffer
against extinction with a VSR. Otherwise, don't bother.
If the animal is acting out of fear, you may be rewarding the fear response. This can happen when you coddle a shy dog.
The timing must be good. If the animal did a great "stay" and you reward after the release, you are rewarding getting up.
The reward has to be sufficient to motivate a repetition. Mild praise won't be enough for some animals. Others require the richest of food rewards, etc.
Reinforcements can become associated with the person giving them. If the animal realizes that he can't get any rewards without you present, he will not be motivated to act.
Animals can get sated with the reward you're offering when they've had enough, and it will no longer be motivating.
Reinforcers increase behavior. If you don't want your animal actively
trying out new behaviors ("throwing behaviors at the trainer"), don't
use positive reinforcement. Use a positive reinforcement to train an
animal to do something.
Negative
punishment is reducing behavior by taking away Something Good. If the
animal was enjoying or depending on Something Good she will work to
avoid it getting taken away. They are less likely to repeat a behavior
that results in the loss of a Good Thing. This type of consequence is a
little harder to control.
The
child has his crayons taken away for fighting with his sister. The
window looking into the other monkey's enclosure is shut when the first
monkey bites the trainer. "This car isn't getting any closer to
Disneyland while you kids are fighting!" The dog is put on leash and
taken from the park for coming to the owner when the owner called (this
causes the unintentional result of the dog being less likely to respond
to the recall). The teenager is grounded for misbehavior. The dolphin
trainer walks away with the fish bucket when the dolphin acts
aggressive. "I'm not talking to you after what you did!" Xena The
Warrior Princess cuts off the air of an opponent who refuses to tell
her what she wants.
Trainers
seldom go to the trouble of associating a particular cue with negative
punishment. It's sometimes called a "delta", from SD or discriminative
stimulus. Some dog owners make the mistake of calling their dogs in the
park and then using the negative punishment of taking the dog away from
the fun. "Fido, come!" then becomes a conditioned negative punisher. My
mom conditioned a similar CP- as "Time to go!".
Positive
punishment is something that is applied to reduce a behavior. The term
"positive" often confuses people, because in common terms "positive"
means something good, upbeat, happy, pleasant, rewarding. Remember,
this is technical terminology we're using, though, so here "positive"
means "added" or "started". Also keep in mind that in these terms, it
is not the animal that is "punished" (treated badly to pay for some
moral wrong), but the behavior that is "punished" (in other words,
reduced). Positive punishment, when applied correctly, is the most
effective way to stop unwanted behaviors. Its main flaw is that it does
not teach specific alternative behaviors.
Our
society seems to have a great fondness for positive punishment, in
spite of all the problems associated with it (see below). The peeing on
the rug (by a puppy) is punished with a swat of the newspaper. A dog's
barking is punished with a startling squirt of citronella. The driver's
speeding results in a ticket and a fine. The baby's hand is burned when
she touches the hot stove. Walking straight through low doorways is
punished with a bonk on the head. In all of these cases, the
consequence (the positive punishment) reduces the behavior's future
occurrences.
Because
a positive punisher, like other consequences, must follow a behavior
immediately or be clearly connected to the behavior to be effective, a
secondary positive punisher is very important. (This is especially true
if the punisher is going to be something highly aversive or painful).
Many dog trainers actively condition the word "No!" with some punisher,
to form an association between the word and the consequence. The
conditioned punisher (CP+) is an important part of training with
Operant Conditioning.
Behaviors are usually motivated by the expectation for some reward, and even with a punishment, the motivation of the reward is often still there. For example, a predator must face some considerable risk and pain in order to catch food. A wild dog must run over rough ground and through bushes, and face the hooves, claws, teeth, and/or horns of their prey animals. They might be painfully injured in their pursuit. In spite of this, they continue to pursue prey. In this case, the motivation and the reward far outweigh the punishments, even when they are dramatic.
The timing of a positive punishment must be exquisite. It must correspond exactly with the behavior for it to have an effect. (If a conditioned punisher is used, the CP+ must occur precisely with the behavior). If you catch your dog chewing on the furniture and you hit him when he comes to you, you are suppressing coming to you. The dog will not make the connection between the punishment and the chewing (no matter how much you point at the furniture).
The
aversive must be sufficient to stop the behavior in its tracks - and
must be greater than the reward. The more experience the animal has
with a rewarding consequence for the behavior, the greater the aversive
has to be to stop or decrease the behavior. If you start with a small
aversive (mild electric shock or a stern talking-to) and build up to a
greater one (strong shock or full-on yelling), your trainee may become
adjusted to the aversive and it will not have any greater effect.
Punishments may become associated with the person supplying them. The
dog who was hit after chewing on the furniture may still chew on the
furniture, but he certainly won't do it when you're around!
Physical punishments can cause physical damage, and mental punishments
can cause mental damage. You should only apply as much of an aversive
as it takes to stop the behavior. If you find you have to apply a
punishment more than three times for one behavior, without any decrease
in the behavior, you are not "reducing the behavior", you are harassing
(or abusing) the trainee.
Punishers suppress behaviors. Use positive punishment to train an animal not to do something.
Negative reinforcement increases a behavior by ending or taking away Something Bad or aversive. By making the animal's circumstances better, you are rewarding it and increasing the likelihood that it will repeat the behavior that was occurring when you ended the Bad Thing.
In order to use negative reinforcement, the trainer must be able to control the Bad Thing that is being taken away. This often means that the trainer must also apply the Bad Thing. And applying a Bad Thing might reduce whatever behavior was going on when the Bad Thing was applied. And reducing a behavior by applying a Bad Thing is positive punishment. So when you start your Bad Thing that you're going to end as a negative reinforcer, you run the risk of punishing some other behavior.
One of the major results of taking away
Something Bad is often relief. So another way to think of negative
reinforcement is that you are providing relief to the animal but of
course, this makes it an example of positive reinforcement - you are
providing Something Good - relief. Confusing?
The
choke collar is loosened when the dog moves closer to the trainer. The
ear pinch stops when the dog takes the dumbbell. The reins are loosened
when the horse slows down. The car buzzer turns off when you put on
your seatbelt. Dad continues driving towards Disneyland when the kids
are quiet. "I'm not talking to you until you apologize!" The hostage is
released when the ransom is paid. The torture is stopped when the
victim confesses. "Why do I keep hitting my head against the wall?
'Cause it feels so good when I stop!" The baby stops crying when his
mom feeds him.
Trainers
seldom go to the trouble of associating a particular cue with negative
reinforcement. You can still go ahead and do it.
Trainers can not control all reinforcers and punishers, unfortunately. There are a number of environmental factors that are going to affect the animal's behavior that you have no control over, but which will still be a significant consequence for your trainee.
Some of
these come from the animal's internal environment - their own
reactions. Relief from stress, pain, or boredom are common reinforcers
and some "self-reinforcing" behaviors are actually maintained because
of this. Examples are a dog barking because it relieves boredom, or a
person chewing on her fingers or smoking a cigarette because it
relieves stress. Drivers speed because it is fun. Guilt is an internal
punisher that some people experience.
There's
actually a fifth possible consequence to any behavior: nothing. You
push the button and nothing happens. You raise your hand and the
teacher doesn't call on you. You get no response to your e-mail, your
proposal, or your job application. The question you then have is, did
no one notice your behavior? Or was it just not worthy of a
reinforcement?
To differentiate between these two possibilities, a trainer can use a no reward marker (NRM). The NRM tells the animal that its behavior will not gain it a reinforcer. A lot of dog trainers use "Nope!" "Wrong!" "Uh-uh!" or "Try again" as NRMs. For example, if you're teaching your dog to sit in response to the cue "sit" (it's not as obvious to the dog as it is to you; after all, dogs don't have the experience of verbal words being labels for actions), and the dog lies down or barks, you can give a NRM. The purpose of the NRM is to get the animal to try something different. It is not a conditioned punisher and should not be used when the dog does something you don't want it to ever do. It's for when a behavior might be correct in a different circumstance but not in this one.
Some trainers also have developed a keep going signal (KGS). This signal tells the animal that it's on the right track, that its behavior is leading to something that will gain it a reinforcer. For example, if you're teaching a dog to roll over and it will lie on its side, you can use a KGS to tell it that it's close to a behavior that will get it a reward, but not there yet.