Reblogged from ClickerSolutions.com because it’s something I’ve been thinking about for a while. Isn’t withholding a treat negative punishment? I’m still trying to wrap my head around it; it feels a bit like semantics right now but… what do I know? I have a lot more to read!
Extinction is the cessation of reinforcement. When you shape behavior, you start by reinforcing a tiny bit of behavior. When you make your criteria harder, you stop reinforcing the lesser offerings. Because these previously reinforced behaviors are no longer being reinforced, the dog stops offering them, choosing instead to offer the new behavior that is being reinforced.
Extinction is not punishment. Punishment is an event. When you punish, you either add something (positive punishment) or take something away (negative punishment) in order to suppress a behavior. Extinction is a “non event.” You didn’t add or take away – you simply did nothing.
Let’s look at an example.
Every day, a person drops a cookie outside a fence for the dog that lives there. One day, however, the cookie lands a bit further from the fence than normal. The dog tries to reach under, but he can’t quite reach it. He tries to dig under, but the ground is rocky. He tries to climb over, but the fence is too high. He tries to go around, but the gate is locked. He will try harder and harder – stretching to reach further, contorting his body for an extra millimeter. He will try a running start to jump higher. But if he never succeeds, he will eventually give up and quit trying. The next time he’s there, he might try again, but he’ll give up more quickly.
This was an example of extinction. The fence didn’t shock him when he reached under or tried to climb. The man didn’t snatch the cookie and take it away. The behaviors he tried simply didn’t work. When one thing didn’t work, he changed his behavior. He tried again. He tried harder. He tried something different. These are characteristics of extinction bursts – characteristics we count on when we increase our criteria during shaping. After he stopped trying to get the cookie, he occasionally tried again. That’s another characteristic of extinction called spontaneous recovery. Had he gotten the cookie during spontaneous recovery, the behavior would have bounced back, very strong. However, with continued lack of reinforcement, the behavior went away again.
So how do you use extinction? When you train, define the specific criterion you are looking for in your session. For example, when you begin training sit, you would reward the dog for putting his bum on the ground. If your dog offers anything else – for example, if he lies down or backs up – ignore it.
Once your dog is reliably offering sits, you’re ready to increase your criteria. Instead of just any sit, you decide to reward only tucked sits – sits where the back feet are drawn toward the stationary front feet. Suddenly some of the sits you used to reward aren’t being reinforced anymore. The dog has to experiment to figure out what will work. This experimentation is part of extinction – try again, try harder, try something different. Gradually he begins to offer more and more tucked sits. Because you aren’t reinforcing non-tucked sits anymore, those sits will extinguish.
Extinction can also be helpful in solving some problem behaviors. For example, a dog jumps on someone to get attention. If people suddenly don’t respond – literally become a statue – the dog will have to try something different to get the attention he wants. If the people instead reinforce sitting with lots of attention – and doggedly refuse to reinforce jumping – then the dog will begin to choose sitting for greetings instead.
However, for this to be true, the trainer must control the reinforcement. If a behavior is either self-rewarding or rewarded by the environment, extinction will fail unless the trainer can consistently offer a better reward. Barking is frequently a self-rewarded behavior. Waiting for a dog barking at passers-by to simply get bored from “lack of reinforcement” is an exercise in futility. When dealing with self-rewarding or environmentally-rewarded behaviors, a combination of management, positive reinforcement, and negative punishment are an alternative solution.
Comparison of extinction and negative punishment:
- Both decrease the occurrence of a behavior over time.
- Negative punishment is an EVENT – the actual removal of something that causes the decrease in behavior.
- Extinction is a “NON-EVENT.” It is lack of reinforcement. Instead of getting something good to strengthen the behavior, or having something added or taken away to suppress it, nothing happens. It just “doesn’t work” to get the desired reinforcement any more.
- Both are frustrating to the learner. The level of frustration varies from learner to learner in each specific situation.
- Punishment is an effective means of changing behavior, but even when used correctly, it may have side effects, including fear and aggression. Extinction has not been shown to have those side effects.
- Extinction causes the dog to change his behavior – try again, try harder, try something different. Punishment, on the other hand, suppresses behavior causing the dog to do offer less and to restrict offerings to what he is sure will result in reinforcement.
- Extinction can cause a decrease in rate of response. However, this is most likely to happen if the overall rate of reinforcement has also fallen, such as when the trainer lumps criteria or increases too quickly.
- Extinction is effective only when the behavior is in no way self-reinforcing to the learner.