the role of punishment

glenmar

Yes, speaking from experience, gsds are VERY biddable, but they are also a little too smart sometimes, and can sometimes out think the human!

Knowing what I know of GSD's, if they anything in the way of smarts as Mini's are, they are very biddable, certainly, but they are also very content to take what you have taught them, and allow them to use it to their own whim and glory!

"Yeah....mom taught me to open the fridge.....now I can finally make the roast turkey sandwich I've been craving"....."This herding stuff is fun, just see how well I can keep these children all in a little row....."

Or in terms of Gaci...."Opening the kennel door was fun, now what other things are my little foothands good for opening?"

Gee, thanks for the warning!

I recently taught Theo to open the fridge but he's not abused it.  Yet.  My fridge is a side by side so keeping a dishtowel on the door is a pain and I just don't do it.  I do put the towel on the handle when I WANT him to open the fridge and cue him to do so.  I hope this one doesn't come back to bite me in the butt!  If I ever have a different type of fridge, guess I'll have to remember the no towel rule!

glenmar

Gee, thanks for the warning!

I recently taught Theo to open the fridge but he's not abused it.  Yet.  My fridge is a side by side so keeping a dishtowel on the door is a pain and I just don't do it.  I do put the towel on the handle when I WANT him to open the fridge and cue him to do so.  I hope this one doesn't come back to bite me in the butt!  If I ever have a different type of fridge, guess I'll have to remember the no towel rule!

Sounds to me like something may have been missing in the foundation of the relationship. In my opinion, a dog purposely seeking out bad behavior is a sign of something more than an "inventive" temperament.

how is the dog supposed to know what is "bad"?  this was a puppy. No one had taught her how she was supposed to behave in a house. Being rather clueless at the time, we made the mistake of waiting for her to be "bad", and then punishing her to inform her that she was being "Bad". Didn't work, since she was capable of finding new ways to be "bad" 24/7. She is what I think of, and why I just laugh, whenever people bring out statements like "the dog can do anything except whatever I told her not to do just now", and when they talk about how teaching dogs to not-do things is an effective training method. She didn't even exhibit the often-seen behavior of "being bad" in order to get attention, either. Once punished for THAT, she stopped doing THAT. But was endlessly inventive.

corgipower

glenmar

Gee, thanks for the warning!

I recently taught Theo to open the fridge but he's not abused it.  Yet.  My fridge is a side by side so keeping a dishtowel on the door is a pain and I just don't do it.  I do put the towel on the handle when I WANT him to open the fridge and cue him to do so.  I hope this one doesn't come back to bite me in the butt!  If I ever have a different type of fridge, guess I'll have to remember the no towel rule!

Oh, he'll figure out how to get the towel, bring it to the door handle and open the door. Unfortunately, he's not likely to figure out how to wash the towels in the process.

Don't worry - I have Aussies.  They'll figure out how to run the washer. 

That was a lot of pages to go through.

I wanted to highlight something Corvus said.

Mild punishment. Mild aversive, if you will. If the behavior is not stopped in the first or second application by whatever punishment or aversive is applied, it is not a punishment, which would cause the behavior to decrease. If, however, a constant application of aversives or things that seem like punishment to us causes the dog to stop a behavior then that is not punishment, either. It is -R. Because, imo, not doing something is also a behavior. The dog has choices. And being quiet is a behavior, at times. It certainly is for me. In a social and/or work situation, there are times I have specifically remained quiet, either to be entertained by what is going on or to avoid social reprocussions. I have stated before that sometimes, my best form of butt-kissing is to keep my mouth shut. It is an intentional behavior to be quiet, sometimes.

Also, I think what gets lost in these discussions is extinction. Behaviors that are no longer offered because they are not rewarding. I think extinction could be paired with -P, or withdrawl of reward. Or the reward of doing what is asked or expected so outshines the other behavior that it is extinguish for lack of a powerful reinforcer, which will vary from dog to dog.

But I do think there are options used as punishment in order to stop a behavior right now. But stopping a behavior right now is not the definition of punishment. Causing the repetition of a behavior to decrease is punishment. Usually on the first application, sometimes second. Any future successful reference to that punishment for avoid the behavior or lighter punishments that eventually achieve the desired behavior are -R.

As MP has pointed out, most people use punishments incorrectly through bad timing or not having a punishment harsh enough to nip it in the bud, as they say. And the problem with a really harsh punishment is that you may not be sure of what exactly you are stopping. You may have shouted or popped or kicked the dog for barking at the doorbell but did you catch the last behavior before the punishment was administered? Was the dog looking at you and now feels punished for looking at you? Dogs can go through several offered behaviors and we have to see them. Barking at the door and looking at you are two different behaviors.

So I'm seeing two definitions for punishment. That a punishment must STOP a behaviour from recurring after a couple of applications and that a punishment need only REDUCE the likelihood of the behaviour occurring again. So in the case of my hare and my dog, doing my pretend "I'm gonna chase you away!" thing when I don't want them into something may or may not be a punishment depending on the definition. It does reduce the likelihood of the behaviour recurring in the future, but it doesn't stop the behaviour from occurring. It turns into a communication: "I'm going to chase you if you don't stop doing that." Being chased is not really aversive to either animal; at times they both quite enjoy being chased, and they both know that if they choose not to run I'll just come up behind them and poke them until they get tired of being poked and move (a nifty trick I learnt from growing up with 3 brothers ). And they also know it can lead to a play session (not so much for my hare, who apparently grew out of play). Anyway, sometimes they don't want to be chased. They just want to be left in peace. So when I come chase them every time they go near a particular object, they learn that if they want to be left in peace, there's a similar object nearby I always leave them in peace to worry at if that's what they feel like doing.

So what is it? The chasing doesn't stop the behaviour, but it decreases it merely because my animals aren't always into being chased and have learnt what behaviours they do make me come and chase them, not because they chasing itself is punishing, really. When they do want to be chased, they can come and pretend to have a go at the objects I always chase them away from and I still chase them, but it's something they want, then. I wonder if it's just communication, the way Penny knows that when I pick up my shoes and her leash, we're walking somewhere, but if I pick up her seatbelt we're going in the car. Or the way Kit knows I'll feed him a raisin if he pokes his nose through his cage at me. So they both know I'll chase them if they hear "Oi" and continue doing what they are doing. 

In terms of operant conditioning, an aversive only has to decrease the likelihood of a behavior to be positive punishment:

"Anything that decreases a behavior - makes it occur less frequently, makes it weaker, or makes it less likely to occur - is termed a punisher. Often, an animal (or person) will perceive "ending Something Good" or "starting Something Bad" as something worth avoiding, and they will not repeat the behaviors that seem to cause these consequences. These consequences will decrease the behaviors that lead to them, so they are punishers. "

http://www.wagntrain.com/OC/

This, for the record, is what I mean when I use the word punishment one these boards.
 

an interesting thought came up between your post, Corvus and that of dog_ma.

As she points out, the definition of punisher is that it decreases the likelihood of a behavior repeating but it didn't specifically give a time frame. I think the reason for the time frame, is that if repetitions of the aversive are necessary, then those repetitions after say two attempts, actually become -R, where the dog learns to avoid the punishment by offering another behavior. That other behavior gets reinforced because it leads to the absence of the aversive.

In your example, how is the undesired behavior reduced? At times, the animals want to be chased because it is play, which, if enjoyable for them, is a reinforcer. Making whatever they were doing worth doing again in order to engage you in play. That undesired behavior at the beginning is their cue to you to play. And you have answered that cue in the affirmative by playing. What you thought was a punishment because of your mindset is not a punishment to them. And the converse, as well. Something I wouldn't have thought as being a punishment is, indeed, a punishment for my dog. A certain tone in my voice. I didn't choose that and discovered it quite by accident. And I wasn't even trying to correct him for anything.

If the def of P is that it is to decrease the likelihood of repetition of a behavior then what do you call something that stops the behavior right now but does not decrease the future recurrence? If you have to do it all the time and every day, whatever you are wanting to decrease is not decreasing. As opposed to +R wherein I have taught a desired behavior in one training session.

I'm not refuting either one of you ladies. I'm saying, let's really delve into this, deeply. You want punishment, you got it. No judgements on each others' words or training styles, no semantics of dialog (I like sunny days. Oh, then you must hate cloudy days).

I think there will be a time frame on punishment because the necessity of a repeated aversive to stop a behavior is -R. Wherein, as a desired behavior is repeated, the aversive is lessened or dropped. In the face of repeated aversives, the dog may change behavior to something desirable as the aversive is dropped, which is -R. Or the dog may continue on regardless because the aversive was not strong enough or is even felt as an aversive, at all, such as collar pops on a GP. And I would think that trying to finger-bite a GP is asking for a bite.

So, then, and remember, we don't get to choose what is an aversive. The dog lets us know by his or her reaction. What is a punishment to a dog? And how long can we call it a punishment in order to be effective as that? As opposed to -R. If punishment is to decrease a behavior, it is not teaching an alternate, desired behavior. If the dog offers an alternate desired behavior and we comply by removing the aversive, that is -R. If we have to consistently apply a correction to stop the dog and they stop but will do the behavior again, it is not a punishment but perhaps a re-direction or interruption or a disengage command or cue. I have a disengage command. But it is not a punishment.

If that behavior of stopping is linked to a command or action that does not actually stop the future of the behavior, then it is a learned behavior and learned behavior only becomes learned by reinforcement.

As for punishment stopping the behavior on the first or second application, it seems like a fair dividing line between P and R, just as if a behavior is learned or repeated after a few applications of rewards, the behavior has been reinforced and is likely to repeat again.

Some interesting questions, ron. P's and R's aren't mutually exclusive. R- requires the removing of something unpleasant, not the absence of something unpleasant. If I correct my dog for jumping today, using P+, then in that instance - when he stops jumping and by doing so stops receiving P+, it becomes R- for the not jumping. If tomorrow my dog remembers that jumping resulted in P+ and not jumping resulted in R- and he chooses to not jump, then he has learned and he may be avoiding P+, but he is not actively in a position to receive R-. He can though receive R+ for choosing not to jump.

If you use what *you* believe to be P, and it stops the behavior at the time but doesn't reduce the likelihood of the behavior reoccuring, then the dog viewed it as a distractor or an interruptor.

I would judge the effect of the P by the frequency of the behavior. If the dog always jumps up to greet people, and always receives what the owner believes to be P, but the dog continues to jump on people at each new encounter, it hasn't been viewed as P by the dog. If he jumps on someone once, receives P, and the next time he encounters a person he doesn't jump, the P had an effect. If he then jumps on someone again, that doesn't mean the P wasn't effective, because it did reduce the likelihood of the behavior reoccuring. So now, on the third encounter, he receives P again for jumping. Now we see if he can meet two people without jumping. If so, the reoccurence of the behavior has again been reduced. Is that making any sense?

With the exception of a serious safety concern, I would not expect P to stop a behavior in one application any more than I would expect a behavior to be solid in one application of R. My dog sits on command once and receives R+. He's likely to sit again on command, but not yet every time I ask for it. Over time he will begin to sit on command more frequently.

I agree with corgi - if an action stops a behavior but doesn't decrease the chance of future happenings, it is an interruption and not a punishment.

Ron, I don't quite understand your point about the number of repetitions and becoming R-? Can you explain? My thought (without understanding your argument) is that there is no time table involved in P+, or even R+. We don't expect a reinforcement to work instantly, nor do we worry about how many liver treats it takes to get a hound to lie down.   Isn't clicker training about a LOT of R+?

I agree that the dog, and the dog's response, dictate what is reinforcing and what is punishing. Our intentions matter very little! I think it is important to know your dog, and to observe how your dog (or any other dog) responds to various stimuli.  

corgipower

would judge the effect of the P by the frequency of the behavior

Exactly, and well put.

corgipower

then he has learned and he may be avoiding P+, but he is not actively in a position to receive R-.

This is where I might differ from you. An aversive is an aversive. If it stopped or decreased the behavior, it can be considered +P. If the next day, the dog changes his behavior to the behavior of not jumping, which is a behavior, as much as jumping is, and you respond by not administering an aversive, then the dog will continue to decrease or extinguish the jumping to avoid the aversive, which makes it -R. The removal or even lessening of an aversive that results in a reinforced behavior. The dog is learning by reinforcement. The dog chose not to jump and avoided an aversive and is therefore engaging in OC.

corgipower

So now, on the third encounter, he receives P again for jumping. Now we see if he can meet two people without jumping.

New situation and new context, therefore the behavior of not jumping is new again. He already learned the behavior of not jumping on you. Now, he must learn the behavior of not jumping others. Different context, new behavior all over again. Dogs don't generalize to the same extent or in the same way that humans do.

corgipower

My dog sits on command once and receives R+. He's likely to sit again on command, but not yet every time I ask for it. Over time he will begin to sit on command more frequently.

With constant corrections, a dog may eventually sit all the time to avoid the aversive of correction, -R. He learns by reinforcement, even if it is negative reinforcement. With positive reinforcement, there is a better chance of leading him directly to the behavior that you want. That is, he doesn't learn from the aversive, he learns from avoiding the aversive. Or, he learns by what is most rewarding.

corgipower

I would not expect P to stop a behavior in one application any more than I would expect a behavior to be solid in one application of R.

I have trained behavior in as few as three clicks. Shall we give the same criteria to punishment? 1 to 3 applications?

ron2

If it stopped or decreased the behavior, it can be considered +P. If the next day, the dog changes his behavior to the behavior of not jumping, which is a behavior, as much as jumping is, and you respond by not administering an aversive, then the dog will continue to decrease or extinguish the jumping to avoid the aversive, which makes it -R. The removal or even lessening of an aversive that results in a reinforced behavior. The dog is learning by reinforcement. The dog chose not to jump and avoided an aversive and is therefore engaging in OC.

Ah. Well, dogs don't really engage in OC. OC is a set of terms that describe a learning process.

R-, as I understand it, is the removal of something aversive. I can sort of see what you mean, but I disagree that avoiding a P+ is the same as removing an aversive. We're into the realm of threats, but I've always thought of R- as an active removal rather than a threat. And R- can't really be used to describe *not* doing something, since R by definition refers to a behavior that is increased. I guess we could loosely say "not jumping" is a specific behavior, but I'm not so sure. Must continue pondering.

Let me put it this way: Pretend we have a dog who jumps, and we're using a shock collar. (Not something I advocate! But using it as an example to illustrate the dynamic). If dog jumps, you briefly shock, and dog learns to stop jumping, that is P+. Even if takes 25 sessions. If a dog jumps, you turn on the shock, and the shock is turned off when dog sits, that is R-. R and P have different target behaviors. I think, Ron, you are suggesting something like dog jumps, shock is turned on, when dog stops jumping the shock is turned off, with "not jumping" being the active behavior that is reinforced. Am I right or did I miss it?

Or: dog pulls against a prong collar. Dog is uncomfortable. Dog stops pulling, "not pulling" or "loose leash" is reinforced through R-. Versus dog pulls, handler delivers a correction. Dog stops pulling through P+, the punishment is directed at the active behavior of pulling.

*off to read more about OC terms*
 

Editing in: it seems avoidance of an aversive is in fact R-. Cool to learn. Back to reading. 

Editing further. This is from wikipedia, which is not a seriously reliable source, but if it is true it a really neat argument for minimizing punishment:

"There are two types of punishment in operant conditioning:

• positive punishment or type I punishment, an experimenter punishes a response by adding an aversive stimulus into the animal's surroundings (a brief electric shock, for example).
• negative punishment or type II punishment, a positive reinforcer is removed (as in the removal of a feeding dish). As with reinforcement, it is not usually necessary to speak of positive and negative in regard to punishment.

Punishment is not a mirror effect of reinforcement. In experiments with laboratory animals and studies with children, punishment decreases the likelihood of a previously reinforced response only temporarily, and it can produce other "emotional" behavior (wing-flapping in pigeons, for example) and physiological changes (increased heart rate, for example) that have no clear equivalents in reinforcement.

Punishment is considered by some behavioral psychologists to be a "primary process" – a completely independent phenomenon of learning, distinct from reinforcement. Others see it as a category of negative reinforcement, creating a situation in which any punishment-avoiding behavior (even standing still) is reinforced."

I would end up quoting your entire post because of all the good points you made.

"I think, Ron, you are suggesting something like dog jumps, shock is turned on, when dog stops jumping the shock is turned off, with "not jumping" being the active behavior that is reinforced. Am I right or did I miss it? "

When the dog jumps and the shock is turned on and the dog jumps off, it is an aversive. The shock collar stopped the behavior at that instant. If the dog, in the future, decreases or stops the jumping It was a +P. But that later process of the dog deciding to not jump to avoid the shock is what I am seeing as -R. As you found out in your later reading.

"Punishment is not a mirror effect of reinforcement. In experiments with laboratory animals and studies with children, punishment decreases the likelihood of a previously reinforced response only temporarily, and it can produce other "emotional" behavior (wing-flapping in pigeons, for example) and physiological changes (increased heart rate, for example) that have no clear equivalents in reinforcement.

Punishment is considered by some behavioral psychologists to be a "primary process" – a completely independent phenomenon of learning, distinct from reinforcement. Others see it as a category of negative reinforcement, creating a situation in which any punishment-avoiding behavior (even standing still) is reinforced."

Exactly. Which leads me to reason that the dog learns through reinforcement, not punishment. And that punishment is not always stopping what you think it is. And that there are other side-effects, sometimes unforeseen during the use, that may happen later. Ergo, punishment does not train. Reinforcement does, even if it's the -R of avoiding an aversive. The lack or lessening of the aversive reinforces the alternate behavior, even if that behavior is not jumping. A dog responds to the environment. The response can be jumping, it can be not jumping. Both are behaviors, imo. As a wise person used to have in her sig, you get what you reinforce.

"d dog learns to stop jumping, that is P+. Even if takes 25 sessions."

As you have learned from your reading, P and R are not mirrors of each other, so 25 punishments in a row are not going to have the same effect or be equivalent to 25 rewards in a row.

ETA:

25 punishments in a row might eventually have an effect of stopping a behavior. But, wouldn't it, by that time, be a matter of the dog avoiding the aversive, finally? Unless, of course, the administered adversive is not seen as adversive. I think, once again, of Dgriego's dog, who can run through brambles and come trotting home with sticky pointy things stuck in his whatsit and never minds it. No matter how many times he runs through that patch. A sharp, prickly pain that most of us would agree is a definite punishment but is not seen as such by the dog or is not aversive enough to override the reward of running through that, whatever that reward is. I know I sound like a broken record. Dogs do what works, even if it requires running through brambles. Another dog might define that as a punishment and not run through there again. And, as you were noticing about what I was saying, the later act of not running through there to avoid the pain of sticky pointy things would be reinforced by the very lack of sticky pointy things hanging from the usually tender part of the body.

From here
"When used correctly, punishment can be reduced to a rarely used, highly effective tool for creating inhibitions. If alternate behaviors are taught with positive reinforcement, the amount of punishment can be further reduced."