I often hear it said that avoiding correction is “withholding half of the information that dogs needs to succeed.” The idea is that by telling the dog both when when they are right AND when they are wrong, that they can learn with more clarity, and therefore be happier dogs who understand exactly what we want from them. You can define correction however you wish.
Is this true?
Not really.
If you teach a dog with clear criteria for success (what you want) then your dog will automatically learn what is wrong, and since being wrong is perfectly fine and is part of the learning process, there’s no reason to avoid it.
Let’s consider it more closely.
On a scale of -10 to +10, with -10 as the most awful thing that you can do to a dog and +10 as being the most amazing reward, what does a dog, or a human for that matter, need to learn?
They need change. A difference. That is all.
So, if you go from a +5 to a +8, then your dog will try to figure out what caused that improvement. Both the 5 and the 8 are positive for the dog, but the dog would still prefer the 8, and will try to figure out what it takes to get more “8’s” from you.
Most of the time, people use “0′ to indicate neutral – the handler will neither reward nor punish, and then a +reward for “Yes, that’s what I want!”
My approach is a bit more positive – I tend to be actively involved in training at all times- whether the dog is correct or not, because I see it as part of the game of training and I believe that it builds our training relationship. So I try to keep all training in the range of about a +2 to +10.
Let’s consider teaching a dog to hold the dumbbell tightly with the following video. How can I teach what is right without telling the dog that munching or dropping it is wrong?
By rewarding what is right at a higher level than what is wrong. I simply need a higher value reward for better work.
Offer the dumbbell and start easy! (that piece is already trained here). Then start moving that dumbbell. If Brito does not clamp down it will come out of his mouth – no cookie. If it stays in his mouth – cookie. Cookie is positive and no cookie is less positive. In the video you’ll see that even when Brito fails I still talk and interact and his tail never stops wagging, so I believe the whole experience is on the positive side of the scale regardless of his success or failure on any particular repetition. He enjoys the game of learning, even when he’s not getting a cookie. That is exactly what I want; no fear of failure.
So how does Brito learn right from wrong? What gets the best reward is right and is most likely to be repeated. What gets a lesser reward (possibly my personal interaction without a cookie) is wrong. No big deal. No need to add a No Reward Marker or any other correction. Just try again.
Three failures in a row? Make it easier. You can see that here, and you can see that this approach allows for success immediately afterwards.
Soon enough, Brito will have learned to clamp down tightly until I request that he release, and soon after that the behavior will simply become a habit – he will clamp down without really thinking about it.
And if sometime down the road he makes an error in a full behavior chain? That’s fine – I’ll hand it right back to him. Because “handing it back” is not the normal flow of the behavior chain, he’ll know that means something wasn’t quite right and he’ll do it correctly on the next attempt. And if he doesn’t? Then I’ll remove that piece from the chain and work on it.
How about under the pressure of proofing? Maybe Brito wants a yummy cookie on the floor so he drops the dumbbell to get the cookie. THEN do we need a punisher?
No. Just pick up both the cookie and the dumbbell. Picking up the cookie prevents self reinforcing and picking up the dumbbell communicates that he blew it, and his chance to earn reinforcement is now being delayed. You can be as cheerful as you wish throughout this entire process.
If your dog understands the process of proofing away from your brand new dumbbell hold then you may well discover that you have single incident learning. One time he’ll drop the dumbbell because of the cookie. You respond by picking up the cookie and the dumbbell. He remembers how proofing works. And that’s that.
Would it be faster to add a correction to really make your point?
No, because dogs that anticipate correction for errors are more reluctant students. It’s not much fun to be told you’re wrong, so the dog’s willingness to engage goes down. And since full engagement is the most important factor for ease of learning, that will not help your long term training at all. Give me a dog that lives to work with me, and training will be a breeze.
There’s no need to add a punisher at all. Not in the teaching phase, not in the behavior chain phase, and not in the proofing phase.
When I was teaching my kids their multiplication tables we did the same thing. First I helped them learn each one individually; we did that by rote. Then I quizzed them; success earned “good!” or moving on to the next one to indicate all was well (maintaining the flow of work is a reward). Errors earned raised eyebrows and a smile to keep them in the game, but the stopping of flow would cause them to reconsider with more effort. Multiple failures in a row? We quizzed too soon; start over and work the individual pieces! Eventually, reduce the amount of verbal interaction for routine success as a matter of the natural raising of criteria -what it takes to get a positive reaction out of me. And finally, stop working on them individually and add them back into their more advanced math that requires the skill of multiplication. And if a problem arises down the road that indicates a lack of fluency? Find the problem area – maybe the 7’s were weak. Work on those separately and then add them back into the whole. Move on.
Train what you want. That’s it.
As always , awesome blog. Thanks for sharing something that is a thorn in the side of new trainers: we need to tell the dog when he’s wrong, right?
I’m continually learning from your input. And I loved this post. But to me… these discussions about corrections *always* end up being a discussion of semantics. I’d suggest conceiving of a correction as ‘anything’ disappointing to the dog… anything less that what the dog is *expecting* or at least hoping for. For people who ‘believe’ that corrections are necessary, this definition can be helpful, as it allows them to apply a correction without the baggage and destruction of adversives. Of course, what is or is not adverse varies with the dog. Brito being, as you say, “fragile”, is intolerant of anything remotely adverse and I congratulate you for your talent for working through that. Some dogs have extremely high tolerances for rough handling and the risk is that applying higher levels of ‘correction’ only builds a calloused attitude and escalate the ‘effective dose’ requirements…. among other issues.
Balance is always the goal and the biggest challenge.
A couple of new concepts have gotten through to me recently. ….. One being recommendations against luring.. interesting arguments there.. the value of the dog solving the problem and learning not to be dependent upon ‘help’ all the time. It can over time build a high tolerance for failure and strengthen persistence in the dog… if done right (always and of course.) I always thought that luring was the shortest path and the value of pure shaping a behavior which could easily be lured had sort of escaped me. At the same time, and by someone else, I’m being told to get in there and ‘make’ it happen using a long line, or grabbing the collar. Provided that the dog doesn’t perceive the handling as too adversive (a tricky evaluation to make), that does seem to help sometimes. It is a bit like luring though. Perhaps classically closer to ‘molding’ … not often recommended. If/when the dog isn’t ‘getting it’ and making progress, surely something has to change and in some situations it is not all that easy to figure out another way to make it easier. Einstein said that the definition of insanity was “doing the same thing over and over again and expecting different results”. So lack of progress? If the problem hasn’t changed, then the dog is insane to continue trying. And if the dog isn’t trying, then the handler is insane to persist.
The other big concept I’m learning is ‘control and manipulation’ of arousal levels. Some of that has to do with relationships, but that’s not all of it. Amazingly I feel that I precision at sub-optimal arousal levels, too low or too high are both compromised. I think I’d have guessed that *any* arousal would undermine precision. I’m finding out that’s not so. I hadn’t understood attitude vs. arousal and the notion of going directly after the level of arousal you want.
As always, thank you for your valuable training insights.
Kathleen
Kathleen, I absolutely agree about semantics. 🙁
Thank you for your thoughts on this subject. I attended a few dog training sessions with a trainer I know, just as a visitor. I like the way she teaches, but what really puzzled me whas when her students were required to the the dog when he is wrong. They had to do it only once they were supposed to know the behaviour (mostly behaviour chains). I just knew that I didn’t want to do this with my dog, but couldn’t put my thoughts together with clear reasoning. Now you did the job for me and I fully agree! I’m trying to think back to those lessons and the engagement of the dogs. It was quite good in most cases, but not great, and some dogs were rather reluctant.
And yes, I did the same thing with “my” kids ( school, teaching) when learning timetables!
I have always considered this statement, that choosing not to incorporate correction is holding back “half the information” to a complete fallacy.
Just look at the things that our dogs learn, and learn very well, that we don’t even intend to teach them and that we have never “corrected” them for. My dogs get up and run to the front door every time my lap top “dings” when it turns off. Apparently, I shut down the lap top and immediately got up and let them out enough times that the ding became a cue – and one of the cues they all know best! They run into the kitchen when I start preparing their meals (but not usually when we prepare our own). They bark with great enthusiasm whenever one of us comes home. They were never corrected for “failing” at any of these things.
This fact alone should, I think, clue us in to the fact that correction is not a necessary step in a dog’s learning. The fact that it can work does not imply that learning cannot occur without correction.
Great topic, Denise!!
Thanks for the video. Are you clicking with your mouth? Please share! 😃
yes.
That was my question also! Thanks for asking!
“Cookie is positive and no cookie is less positive”. No. Cookie is positive, and no cookie is a negative; in the parlance of Skinner it’s a negative punishment. If you are going to make claims based on beginning science, you should at least be consistent. Skinner’s language is already unclear and counter-intuitive; not being consistent with it makes it even harder for people to understand what is a correction and what isn’t.
As for withholding half the information, sorry guys but it is true. The disconnect here is that for some reason you all seem to think a correction has to be not only a negative, but a big awful negative thing you do TO the dog. In the real world, where words mean what they always have meant, doing something TO a dog is called punishment, not correction. A correction is what a trainer does to help the dog be right, such as withholding a cookie when the dog makes a wrong choice. “dogs that anticipate correction for errors are more reluctant students” are not dogs that are being corrected, they are being punished. There is no reason for a dog to be reluctant to get a correction.
The problem that one would run into with the idea that “no cookie” = negative is that a dog who operates with that mindset will shut down altogether when “cookies” are no longer forthcoming for every correct repetition.
Obviously, good +R trainers are not using “cookies” in a way that conveys that message, or else we would not have successful dogs earning Q’s and titles and championships in all manner of sports. Take a Rally FrEe course, for example. The dog is performing 15 different behaviors, many of which are highly complex, at a high level of precision. The handler may have no food on his or her person, and all behaviors on verbal cue is desired. And, in between each of the 15 behaviors, the team moves together with the dog heeling on either the right or the left. The quality of this heeling is scored on attention and animation (so we work to cultivate joyful heeling).
If the message being sent to the dog in training were “no cookie” = negative, we would be sending the message “that was not correct” to the dog 15 times, in addition to the heeling.
So, it stands to reason that we need to move beyond Skinner’s work when considering the use of reinforcement in training. At some point the dog needs to learn that “no cookie” can actually mean, “you’re correct, off to the next task!”
I would agree with Denise’s assessment that in a context where many different reinforcers are available, “no cookie” (but keeping up a cheerful attitude, offering praise, etc.) is not as reinforcing as “cookie”. But “negative” . . . that would present major problems down the road.
This is something that I learned the hard way with the second dog that I set out to train seriously. I used treats in a way that conveyed, “cookie” = correct; “no cookie” = not correct. And when we started competing, without me having food on me, it all fell apart. I have learned since then how to use treats in a way that conveys, “cookie” = correct, “no cookie” = nothing in particular (and then, from there, I can build behavior chains, duration, etc). I am very careful now never to use treats in a way that conveys “no cookie” = not correct.
Good use of reinforcement is quite an art unto itself. It’s fascinating.
I don’t know what Rally FrEe is, but if it is like AKC Rally, you are allowed to use several commands, screw up one or two stations and essentially beg the dog thru the course, and still pass. So that isn’t really a trained dog.
WRT no cookie meaning the dog is wrong, and it carrying over into the ring, that is the same thing that any trainer has to deal with. If the dog is still dependent on feedback – of any kind – it is not ready to go into the obedience ring. He needs to know what is the right response to all commands, and that any other response is wrong, without getting a cookie for the right ones or thinking that the lack of a cookie means he was wrong.
This will be the last response posted on Kathy’s thread.
A new trainer can consider the phrase “is avoiding correction withholding half of the information” and can watch the training of the person who has told them that. They can see how that person defines a correction – what it actually looks like. Then they can consider that phrase against what I show in my video, to have a clear visual representation of what I mean as well. We don’t have to worry if your definition of reinforcement is the same as mine or if your idea of a correction meets my experience. I don’t argue semantics – I show video examples. I haven’t found that arguing semantics progresses my training any, so I don’t bother.
New trainers are hardly capable of understanding what they see.
The video was indeed valuable to this new trainer. I think I AM capable of understanding what I see! Semantics aside, I was able to see exactly what I was doing wrong in trying to train my dog to hold on to an object for a longer period of time. I had been trying to act “neutral” when my dog failed to hold the object, by ignoring him. Whereas if he did hold on long enough for me to click, I’d praise and reward. Unfortunately, after a few ignores, my dog would lose interest in the game entirely. This video showed me how I can continue to be enthusiastic and make the activity fun, while holding out that higher level of reward for holding on to the dumbbell. You can say the semantics are wrong, but the video is pure gold!
When you are teaching a new skill, there is NO wrong. Only a ‘not what I wanted’. All tat is need is to let the learner know that what s/he just did is not what you wanted. Withholding the big payout should be enough. and giving an encouragement reward (aka lesser reward) should keep the learner still engaged with the task.
“Correction” technically should be showing the learner what you wanted instead of the “incorrect” response. So technically we often ‘correct; I training. Mostly by going back to a earlier stage of learning, but also in determining just why dog is failing and then changing our training tactics.
Punishment technical is used to stop an unwanted behaviour. And I use it often — though it usually only works when the dog/boy/girl/cat is caught in the act. Which means that in psychological/behaviorism terms I am only using an interrupter since, while the behaviour stops at that time, it does not seem to decrease in frequency 🙁
I just saw the video, I love your style of training! Happy wags the whole way through.
This reminds me of how I taught Matilda her awesome fetch skills. First, I taught her to drop things into my hand by giving them to her, and praising her the moment it fell out of her mouth and into my hand – Then, I started throwing her ball, and rewarding her if she brought it into my hand. If she dropped it near me, I’d just toss it again. That’s all she needed to show her that only a complete retrieve with the item in my hand was correct.
I also tell her, “Wow, you’re so smart! You’re the smartest!” – I think a dog can take a compliment! 🙂
Sheesh guys – the semantics are built into her scale. that’s what’s so lovely about it. of course giving a +5 reward instead of a +8 is a correction. but the point is that it’s on the positive side of the scale (as a conceptual device….). and different dogs are sensitive to different amounts of negative. My Tamaskan hates to be wrong – so I have found i get much better energy from him by being really positive (still no cookie…lol) and just trying again…and it’s really not good if i set the parameters so that he fails more than once. but my friend has a lab…that dog hardly seems to care!
A +5 cookie isn’t a correction, it’s an indication to the dog that he needs to try something different or try harder IF he understands that this reward means less than that reward. A correction is specific to whatever the dog is doing wrong; if he breaks a sit you take him back and have him sit again. This isn’t semantics, it seems to be people who don’t understand what a correction is, or maybe that each dog needs a training program tailored to his temperament and current mood?
To avoid pointless debate, this will be the last comment on this particular discussion thread. The topic of this blog is whether a dog needs corrections to understand what is right. If you need clarification about what I mean, please watch the video and read the blog – I have certainly attempted to be clear.
Reblogged this on DogSentials and commented:
This is an excellent blog post by Denise Fenzi on why corrections are unnecessary and counter productive.