These days, many trainers eat and breathe learning-theory. The Little Book Of Learning Theory (LBOLT) is creedally accepted by one and all, even though little of it works in practice. Please don’t stone me. I am not being heretical. Learning theory is a real and valid description of how computers train animals but the LBOLT offers little for when people teach people or train animals. People are not computers and have neither the consistency, computing power or timing. LBOLT has many constraints in practice. To make matters worse, the really useful principles that LBOLT has to offer are ignored by many trainers, for example, that consequences are binary — from the dog’s point of view either things get better, or worse. Instead, trainers will debate for hours, which quadrant they are in, even though it’s all really a moot debate. A trainer might consider a procedure to be negative punishment (time out from play) but the dog perceives it as positive punishment (solitary confinement in prison watching other dogs play).
During my years at UC Berkeley, I was the only PhD candidate who had never taken a Psychology course. I was surprised that so many students accepted learning theory as gospel and assumed that the principles would work when applied in human education, motivation, or animal training. Because of my lack of a formal psychological education, I accepted nothing and questioned everything, especially everything to do with punishment. My recurring thought was, “that might work in a laboratory when a computer is administering shock and the rat has no choice, but that’s’ never going to work in real live when people are trying to train animals.” Try that on a dog and either the dog will just terminate training by running away, or convince the “trainer” to run away. (Of course, if the dog does the latter, he will probably be killed.)
From growing up on a farm and always being surrounded by animals, I realized that most punishment doesn’t work in practice. Because, unlike computers, people are outrageously inconsistent. The animal (or child, or employee) first learns those times and situations when misbehavior is possible without punishment, i.e., when the cat’s away the mice will play. When designing the Animal Behavior Department for the San Francisco SPCA, I realized that we were asking the wrong questions. Rather than asking, “How can I punish my dog for misbehaving when off-leash, or when left at home alone?” we should be asking, “How can I reward my dog for behaving appropriately when off-leash, or when left at home alone? Whoooo Hooo! Light-bulb moments. Integrate short training interludes into off-leash walks and play — sit/go play, sit/go play, sit/go play etc., and the distraction becomes a reward. “Go Play” may now be used as a reward to reinforce sitting when requested. Similarly, stuff chewtoys with food and each piece of food rewards the dog for chewing his chewtoy and for lying down quietly and calmly. Brilliant. We just eliminated owner-absent destructive chewing, excessive barking and hyperactivity simply by feeding dogs from chewtoys instead of bowls.
So laboratory punishment schedules don’t work that well in dog training without adhering to a number of strict criteria. However, after I started teaching Puppy Classes in the early 80s, I quickly realized that most schedules of reinforcement don’t work that well in dog training either. Differential reinforcement is the only schedule that makes any practical sense in terms of maintaining and increasing frequency and continually improving the quality of behavior.
Over one hundred years ago, the study of animal learning was whelped from dog training. Before Edward Lee Thorndike started his landmark learning studies, he asked local dog trainers how they trained dogs. For 90 years dog trainers pretty much ignored learning theory. Joel Walton repeatedly suggested that I titled my 1980s learning theory seminars, “Learning Secrets” because at that time, learning theory a secret for most trainers. When dog trainers embraced learning theory, pet dog training took a quantum leap and became science-based dog training.
However, nowadays the LBOLT is literally worshipped and dog training is in danger of changing from a single science to differing religions with sectarian fundamentalist trainers only worshipping and interpreting their half of the LBOLT. There seems to be more discussion and expression of emotional views and feelings about what is right and what is wrong, than there is about quantifying quicker, easier and more effective ways to change the frequency and quality of behavior and hence, the dogs’ and owners’ quality of life.
I think it is time to come full-circle and for pet dog trainers to rewrite the LBOLT as a simple and practical text.
The first gift that we can give to all animal owners, parents and teachers is to simplify the ridiculously ambiguous and unnecessarily complicated and confusing terminology. Second, let’s simplify the underlying theory by going back to Thorndike’s original premise — that behavior is influenced by consequences and that consequential feedback is binary. Thorndike’s Law of Effect stated, that the frequency of a behavior is dependent on its immediate outcome: if the outcome is pleasant, the behavior will increase in frequency; if the outcome is unpleasant, the behavior will decrease in frequency. Third, rather than unsuccessfully trying to mimic consistent computer quantum feedback with kibble and shocks, let’s train dogs off-leash and use our voices as feedback — praise and instructive reprimands. We cannot mimic computers but we can make sophisticated, inimitably human, qualitative judgments and we can give more varied and instructive feedback. When we use verbal instruction and verbal correction, what we do simply cannot be defined by existing computer-generated learning theory. The LBOLT remains a black and white cave painting compared to our Avatar.
To hear more of Dr. Dunbar's views and insights see him live at one of his upcoming seminars.
- dr. ian dunbar's blog
- login or register to post comments
- 593 reads
- back to the dog blog
- more by dr. ian dunbar
The Law of Trichotomy
The Law of Trichotomy states that two things can have one of 3 relationships: one is bigger/better/more than the other, the opposite, or they can be considered equivalent, the same.
In practice, I observe the Law of Effect to have a third effect: What we identify or intend as a "consequence" can be termed by the dog a "non-event", a zero, equivalent to every other time. If the outcome is deemed a non-event, the behavior is likely to continue with the same frequency.
All nerdiness aside (am I even capable of that?), having 3 grades for a performance consisting of so many components as an organism presents is, I think you are correct, rudimentary at best. Qualitative and varied verbal feedback, with its nuances and subtle shades of color, provides a distribution of "rightness" or "wrongness", a continuous scale, that allows us to convey that some repetitions of some behaviors are some slight degree more optimal than others, rather than the black/white, sufficient/insufficient, yes/no, angel/demon ideas used by raw theory as it stands now. Adding more content to our communication with our dogs can only enhance the precision with which we can work.
The relaxed, almost nonchalant "no", LOOK, LEAVE IT or COME command as puppy approaches the favorite off-limits chewing target in the house is vastly different from the panicked NO!!! COME!!! command as adolescent dog trots toward traffic or other danger. And in these variations of tone, and all the situations in between, there is not only information, but meaning to be found by the domestic dog in a way that perhaps no other species is equally capable.
And, I would be remiss if I didn't note that we need research on this, too. =)
I had to Wiki The Law of Trichotomy
Hi Nicole ... Thank goodness for Google and Wiki. I get it ... if we have x and y, then 1. x>y, 2. x<y and 3. x=y.
So, with binary feedback (rewards and reprimands), we can construct another quadrant in terms of their binary usage.
1. Ignore the good and moan and groan at the bad — this tendency is, of course, ubiquitous in every aspect of our lives and is most certainly the worst of our human foibles. For years it dominated dog training: Ignore desired behavior and reprimand /punish the dog for undesired behavior. Takes forever to train the dog, that is, if the dog sticks around for "training".
2. Ignore undesired behavior and reward the dog for desired behavior. Very popular and effective in all-or-none reward training and clicker training. Fun way to train but often takes a long time to reach criterion.
3. Reward the dog for desired behavior and punish the dog for undesired behavior. Now this, traditional binary feedback, gives twice the amount of information that the previous two and so, dogs learn quicker. BUT so many trainers do not like this because of the assumption that punishment has to be aversive. (Because punishment WAS always aversive in learning theory experiments — punishment had to be shocking because the computer could not clear its throat or sigh in desperation.) However, when we replace an aversive punishment with "nope" — information that you have made the wrong choice, things start to speed up. And when we use our voices profitably and instructively correct the dog's behavior, training speed takes off like a rocket and the dog learns lightening fast. In a sense, the verbal instruction/correction is a reprimand, hence my term — instructive reprimand, even though it is given in a soft voice. I think that we should think of correcting dogs along the lines of how we would teach a child to read or throw a baseball. When a child pronounces "cat" as "bat", it's wrong and if we would like the child to learn to read, we should let them know it's wrong and then reaffirm what is right. "Ahhh you use a bat to hit a ball. B-A-T is bat. This is cat. C-A-T spells cat. Miaow miaow. Kitty cat." And when the child pronounces cat as cat and bat as bat, it's time to celebrate. Whoooo Hooo! This is the type of binary dialogue that we are missing in dog training. Trainers are becoming machines and so, dogs are becoming machines and dog training is in danger of losing its soul.
4. And then we have the non-event, ignore bad behavior and ignore good behavior. Sounds pretty dumb right. I mean who would do that? Well ... most dog owners, parents and employees for a start. We have to get people talking to their dogs again and then perhaps they'll talk more to people.
Good Dog, Bad Dog..
Chico is my 1st dog of my own, Boxer x Staffordshire Bull Terrier and I decided - before collecting him - that I would not be using the choke chain/rub the nose in pee approach my dad taught me to train our family dogs with (Dobermans).
I knew nothing about training. The only 'trainer' I had ever seen came to our and shook a bottle (filled with coins) at our rescue Doberman. Sasha had been confiscated from her owner at 6 months old, as a severe cruelty case, and was prone to (what appeared to be) random & spontaneous aggression. She bit the trainer and every other member of our family (breaking skin multiple times). So, my dad was told to 'dominate' her. What came next made me cry for weeks. You wont be surprised to learn she was aggressive until the day she died, 9 years later. Although my siblings & I convinced our parents to keep her alive, she was dangerous and (looking back now I see) she was often unhappy. I felt like I had failed her. There was no way anyone was going to convince me to see a trainer, or go to a class back then (9 years ago)!
So, I pretty much trained Chico using the only method I could think of (age 21 years old) that was essentially the 'good dog, bad dog' approach. No force or physical contact. Just words backed up with good consequences (food, fuss, petting, games), no consequences or bad consequence (sounding mildly cross, sending to bed, refusing game). Socialisation came naturally, because I lived in the city centre and led a busy life. Wherever I went he would come along and I made new experiences fun for him. I spent a lot of time saying good dog/bad dog back then and the good:bad ratio grew rapidly.
Chico's never going to win any obedience awards, but he is a fabulous pet. I can't think of a time he's ever tried to push a boundary. Never felt the need to guard resources. He's a good traveller. Has good recall. He likes fireworks, motorbikes, people, dogs....... He communicates clearly to me if he scared/happy/sad/unwell etc. and I respond to that as required :)
Now I guess him turning out so well could be just pure luck. Or maybe the easy good dog/bad dog approach allowed me to be more consistent?
I think a verbal signal to say 'nope that's not right' is a good tool, as long as there is plenty of signals to say 'yes, that's the right thing' backing it up. Seems like common sense to me, but I'm not a trainer ;)
remembering 3 things at a time
...I'm wondering if I can add what has helped me if you're looking for ideas on what regular people like me found helpful in classes.
I'm not a scientific mind. I'm pretty uncoordinated, but I caught the lure/prompt/capture/shape get happy bug.
I got hooked because the trainers fed me a little bit of info at a time, just enough to get me hooked and cheer me on when I accomplished a baby step, so I couldn't wait till we had our next class. Simple as that.
My coles notes that help me in 3's
A: Remember my part;
1- use a hand signal that pretty much lures them right into position so not much guessing is involved.
ie. the lure, stinker the better at first to make things really a no brainer. Or if shaping..have a mental picture of the baby steps that it will take to perform the behaviour.
2-know what the rest of my body supposed to be doing.
ie.. learning 'come" run away and call your dog to you...don't lean forward.
3- know when I'm supposed to "mark"/capture/reward the behaviour and party.
ie. I use "Yes" but noticed that some sounds naturally make my dogs bum wriggley (WOW, Awesome, Superstar) so I also use those after "yes" as a bridge until I can get their reward to them. That way even if my timing is off...they get to know the party's still on. Be genuine about partying with your dog.
B. Helping them succeed: Environment & what to do when something doesn't work.
Things for me to keep in mind that speed up the learning process:
1-look around and remember that their hearing, sight (dichromatic vision) of quick movement and nose can distract them and kibosh what I'm trying to communicate to them..so start something new in the quiet of your home. Then test it and start all over in a new spot adding little bit of distraction at a time.
2-You're the magician and their cheerleader...you can make all the things they love to do and eat and play with appear...mix up the rewards to keep things interesting. Be a part of that party. It will do your health a world of good if you let the party flow into the rest of your life!
3-If something goes sideways, I'll either redirect them into an incompatible behaviour, to a place so they can "reset" or I just ignore whatever happened and then go back to the drawing board before trying again. Just remember to keep things short! 10 sec - 2mins tops...and party inbetween.
C. Always remember...
I found Bob Bailey's quote really helpful after learning the basics of luring and shaping.
Bob Bailey's "...let the animal tell me what behaviors it had. I then (prompted)(shaped) as needed to get the behavior as quickly as possible. To change an animal’s behavior quickly you should change your behavior ASAP. Doing more of the wrong thing does not make it right....
happy-houndz.blogspot.com cheers, kate
I think there needs to be a
I think there needs to be a bigger distinction between trainers who are "becoming machines" due to not talking to their dogs at all and only click/treating, and those who use clickers and don't use no-reward-markers but DO talk to their dogs, praise their dogs, and get excited with their dogs. I wish those of us who use clickers, shaping, and learning theory while having fun, getting excited, being verbal, and playing games wouldn't get lumped in with those who are machine-like and non-verbal in their clicker training. Not that I'm convinced there's anything wrong with that approach, but it's not how all (or even most in my personal experience) clicker trainers train.
I am confused about this section: "rather than unsuccessfully trying to mimic consistent computer quantum feedback with kibble and shocks, let’s train dogs off-leash and use our voices as feedback — praise and instructive reprimands." Is this really an either/or scenario? Ignoring the shocks for the moment (probably not many shock trainers on DSD), don't praise and instructive reprimand have to be linked with reward/no reward to mean anything to the dog? Otherwise it's all just noise. Dogs can most certainly learn to love praise, and learn that those eye rolls and sighs and "nopes" mean he didn't behave correctly, but it is my undertanding that to teach these verbal concepts a trainer must first give them meaning, and make sure they retain their meaning ("no" will quickly lose it's meaning when there's no consequence connected with it, in my experience at least).
If instead, what you mean is get rid of kibble and shocks and replace them with giving and taking away access to life rewards, then we are back to common ground. I and most clicker trainers I know use click/treat to train a behavior, and life rewards to maintain it - otherwise you'd have to carry treats and clickers for all time and no student is going to agree to that. But I don't want to put words in your mouth, and I was unclear on what you meant the alternative to be. :)
Not quite what I meant with Law of Trichotomy
So, the LBOLT is the place to start, because of it's binary model, which is better than a lot of other places to start like, "My parents always did this," but when it comes to training domestic canines, binary isn't the place to stay. I think it is difficult for many people to accept the (apparent, from the perspective of the new human handler) "slop", because "it doesn't look like training". Training means carrot or stick. Kibble or shock.
The complication comes in because although carrot/stick is binary, the learner's viewpoint is not binary, but trinary. My behavior was good, acceptable/ineffective, or bad. The dog was rewarded, punished, or neither. The dog was made happy, disappointed, or stayed about the same. This model is the next step for humans in learning about and evaluating training. Carrot, stick, or neither.
When we don't intentionally indicate all of the three possible outcomes (scores, not kibbles/shocks) to the dog, that third outcome, the zero, the no reward or no punishment becomes confusing. Dogs don't know what quadrant they are in! And, in real life learning events, not artificial set-ups, the quadrant is always changing. When nothing happens, is it a non-reward (punisher) or is it a non-punishment (reward)? By providing feedback that indicates correct, incorrect, and "don't worry about it" (mulligan), we don't have to force the dog to guess, he KNOWS the nothingness is a mulligan, very little to be learned from this repetition. Humans are uncomfortable with not rewarding AND not criticizing (punishing), but simply observing, and letting go. Perhaps it makes us feel powerless. And perhaps that is because, in fact, we are. I think the matter at hand to move forward (or backward?) to "soul", to real, substantial social interaction with canines as the accepted definition of "what training looks like". Equipping people with the knowledge and skills to look for what is easy to overlook, because, in a sense, it isn't there.
If I understand you correctly, what you call the "soul" of dog training is providing a far less clear-cut range of outcomes (grades, judgements, scores) using not only the words or markers, but emotional content, a range of tone I find myself at a loss to capture in text. This reality would be the goal, the last step of the learning process, from screaming or violence or standing helplessly to having the pet dog we want. The polar opposite of computerized. Vibrantly present, appropriately interactive. Sharing genuine experience, rather than doling out one of two outcomes, the metaphorical "kibble" and "shock".
The reality of performance is that there is a gradient from black to white, not all correct and all wrong. The range is partially captured by the calm thank-you-keep-doing-that "good", the surprised/excited "GOOD!", and the slow g-o-o-o-d we use for a sit stay. And there are a whole range of in betweens. That third outcome, that non-event, that middle ground can be closer to correct or closer to wrong. There's the matter-of-fact oops "no", and the you-will-not-be-doing-that-again "No." (My mother used to say, "No. N. O. That spells 'no'." The whole phrase.) There is the excited command issued as a reward, and there is the consequence command. "You miserable hound!" is what one says when a muzzled rowdy type disappoints. I can see and hear you doing it, Ian, on a video I watched some years back, my 1st non-textual contact with you. Hilarious, but also very real, very genuine.
This range of most correct to least correct provides significantly more precise feedback to the dog, provided the variation in human emotion is intentional, and not reactive. (Well, even when the human IS reactive, it tends to tell the dog, "Hey, you're on your own. I've lost control even of myself...") However, it is difficult for new trainers/handlers to pick up on one tone or another. Newbies are too busy looking to identify only rewards or punishments, and, unfortunately, they have a (rudimentary) idea of what they thnk they should be looking for: presence of choke chain=punishment, even if it is functioning quite humanely in a cue capacity, presence of food=reward, even if it's being jammed into a reluctant dog's jaws (yes, this is punishing), alpha roll=pure unmitigated evil, clicker=pure creamy goodness, or vice versa, depending on your background. People are significantly more affected by straightforward visual cues, still images of the interaction, than by the functional and social components of an interaction, including the pace, the direction, the timing, the tone. And, certainly, presence or non-presence of items or "techniques" is initially much more distinguishable than analysis of the dog's perception of the item's or technique's functionality.
As trainers coaching humans, the situation becomes even more complicated, because we look at an interaction and see human correct + dog error, human error + dog correct, both correct, both error. And, of course, a whole mix of faults that would take a team of lawyers to sort out intent, malfeasance, etc.! For each of these combinations, the combined performance can result in correct, incorrect, or some mid range outcomes.
I'm quite sure I'm still failing to capture the ideas I'm presenting. Is this at least in the same vein as what you intend when you talk about "soul"?
Training is ALL about teaching dogs ESL
Hi Meghan. Really good comment. Thank you. Please don't misconstrue what I am saying. I love lure/reward and clicker training. After all, I was the person who reintroduced and popularized the use of food lures and food rewards some 30 years ago. Also, if my memory serves me correctly (I'll have to ask Karen here), years ago Kathleen Chin and I hosted and promoted Karen Pryor & Gary Wilkes's very first doggy clicker seminar/workshop. Using lures and rewards, especially food lures and food rewards are essential in pet dog training. However, we should not forget why we are using them — so that people can communicate to their dogs. The very first stage of Lure/Reward Training involves phasing out food lures, or in other words, teaching dogs ESL — doggy words for dog behaviors and actions, so that owners may now instruct their dogs as to what they want them to do and calmly and patiently instructively correct them whenever they err. Food lures and rewards and clickers are essential training tools, but they are meant to be temporary and be used for a purpose, not forever.
Training comprises generally increasing or decreasing the frequency of behaviors and teaching a reliable presence or reliable absence of a specific behavior on cue. Training animals is about teaching cues — words, so that we may communicate with them. Not words like "No" or "Arrrrghh!" which simply let the dog know that he is doing something wrong. ("What else is new?" says the dog.) But words like, "(get your) Chewtoy", "Outside", "Sit" and "Shush" words that communicate three things: 1. You're making a booboo , 2. The severity of the transgression (urgency of tone) and 3. What I would like you to do (the meaning of the word).
Dog training is all about teaching animals the meaning of words, so that we may use them to communicate. There is nothing wrong at all with using tools to do this but surely we don't want to limit our communication to kibble and clicks and jerks and shocks for the rest of the dog's life. Surely, there's a point (as soon as possible) to phase this impersonal quantum feedback (that was necessary to teach the meaning of words) and then start to use words (that we have taught) to talk to our dogs to give them clear instructions and clear and rich understandable feedback.
"Yes!"
Nicole, laughing (happily), I just wanted to answer your last question, "Yes!" Thank you for an insightful comment.
LBOLT is B&W and it's intentionally B&W to facilitate dogs understanding instruction and learning how to read us (and so, not get into trouble). B&W is essential in terms of giving direction, but I think dogs deserve more in terms of feedback. I would like them to get 3-D, technicolor Avatar feedback, so that they can learn to read our every nuance. I want people to have a vibrant relationship with their dogs and I want people to learn how to read their dogs, so they truly understand their needs, their loves and fears, and their preferences — what they like and what comes next..
I love science and I love clear feedback. But it saddens me when the feedback is always so formal. It's like a father who gets his secretary to send his son a check for his birthday. Just a couple of words would be worth so much more. Again, dog training is about teaching dogs the meaning of words so that we can use them to communicate.
As an aside ... just this evening I learned something, that, although hardly world-shaking, blew my mind. Kelly told me that Hugo needed to poop. I took him out front and he peed and then peed a second time. Kelly told me that if he's pees twice out front, he really needs to poop out back. I took him out back and he pooped within seconds. I researched dog pee/poop behavior for ten years but Kelly read Hugo like a book.
Wonderful!
The first gift that we can give to all animal owners, parents and teachers is to simplify the ridiculously ambiguous and unnecessarily complicated and confusing terminology. Second, let’s simplify the underlying theory by going back to Thorndike’s original premise — that behavior is influenced by consequences and that consequential feedback is binary. Thorndike’s Law of Effect stated, that the frequency of a behavior is dependent on its immediate outcome: if the outcome is pleasant, the behavior will increase in frequency; if the outcome is unpleasant, the behavior will decrease in frequency. Third, rather than unsuccessfully trying to mimic consistent computer quantum feedback with kibble and shocks, let’s train dogs off-leash and use our voices as feedback — praise and instructive reprimands. We cannot mimic computers but we can make sophisticated, inimitably human, qualitative judgments and we can give more varied and instructive feedback. When we use verbal instruction and verbal correction, what we do simply cannot be defined by existing computer-generated learning theory. The LBOLT remains a black and white cave painting compared to our Avatar.
Brilliantly said, Dr. Dunbar. Thank you.
Why does trainning have to be work..
I have been interested in Dog training for a few years. I have noticed that the most reliable habits my dogs learn are self taught. IE bed time When i get up and go to bed all my dogs line up at my bed room door. They some times will go their if it is around 10 pm or if i go to the tv and it is late. But They always go their if i turn the tv off late at night even the new dogs learn this fast.
Last summer i started to use dog on a rope with all my new dogs. For a few weeks i would leash the dog to me and do what I do go to store walk around looking for my keys, house work ect. I do not talk to or encourage or discipline the dog. Just went about my business. by the end the leash would fall off because it was slack. the dog would watch me and follow in a calm manner and stop when i do. When i do open communication with the dog it is receptive and a little attention starved and watches me. Formal training is very easy after that point.
I believe that though this lesson the dog is learning to read me and we develop a relationship based on trust and respect.
I have also noticed that many poorly behaved dogs when removed from their home environment and live with me and my dogs will drop many of the habits or they become less pronounced almost from the moment they leave the old environment. with out correction or overt training. So that makes my mind think that it is far more important how we act and behave around dogs then actual training the dog.
This i believe is the same principle that is involved with the bed time ritual.
then again i could be very wrong LOL
one last note when ever i loose my keys all my dogs follow me though out the house like it is a training game I wish they would just find them so we can go for a drive
I have not had a puppy to train in many years and deal mostly with aggressive or very timid adult dogs.