This conference has eighteen participants: three other people from Leader, two from Pilot Dogs, three from Guide Dog Foundation, one from Canada, one from Switzerland, three from South Africa, two from Australia, and two from Finland. So not only do we get a huge amount of information and instruction, we also get to meet GDMI's (Guide Dog Mobility Instructors) from all over the world. You can see why this is so exciting!
Guide Dogs for the Blind is the school that I first started raising puppies for, way back in 1988. If I hadn't been a puppy raiser, I never would have known that this was what I wanted to do with my life. So it is always nice to come back to GDB for a visit. They have a very beautiful campus (as do all the schools I've worked at, really) and in appearance it has barely changed at all over the years, although this is my first visit since they removed the big outside stage where graduations used to take place. The weather is perfect here this time of year -- classic California sunny (but not too hot) with a sea breeze blowing all the time. Walking through the beautiful landscape to the beautiful new Assembly Hall was like walking through Paradise.
The conference began with an overview of why and how clicker training was incorporated into GDB's training program, beginning with limited use of food reward in 1999, the same year I started working at Guiding Eyes. Then it covered the science of positive reinforcement training. Michele Pouliot is the main presenter. She started working for GDB just before I was born and has been, as far as I know, the single person most responsible for promoting the use of positive training methods for guide dogs. She is also a nice person. What I mean by this is that there are a lot of positive trainers who are so contemptuous of traditional training that they immediately turn people off. Michele was a traditional trainer for a long time and does not appear to have forgotten any of it. She really is good at teaching this stuff and answering questions in a totally non-judgmental way that makes her easy to relate to for just about anyone. There's really nowhere else in the world that a GDMI could get expert instruction in clicker training with such a narrow focus on guide work-specific skills. Really this conference would've been cheap at twice the price for the knowledge I got out of it.
After explaining the history, Michele then went on to explain the science of clicker training. Clicker training is really just a type of marker training, which is a type of operant conditioning. Operant conditioning works through using reinforcement and punishment, both positive and negative, to alter behavior. While you don't have to turn into a science geek in order to use the clicker to teach guide dog skills (which is good, because I never will), you DO have to know the science behind it so that everyone within an organization is using the same terminology when discussing training. Also, everyone has to get good at defining goal behaviors and breaking them down into their essential elements. So, for example, if someone asks me what is the goal behavior I want in a guide dog and I answer "Focus", that doesn't really tell anyone much. To implement this, or any, structured training program successfully, there needs to be both precisely defined behaviors and some sort of standardized outcome measures.
This discussion also reminded me of some things I had already heard in previous clicker training sessions, both the earlier one Michele gave at Leader and the even earlier ones Emily and Sheila gave. One of these things is that it really is necessary to have a very high rate of reinforcement when first teaching a new skill. I know this, of course, but I also think back to the number of times I have been standing there with a dog and a chair and waiting… waiting… waiting… for the dog to nose target the chair while time ticks by on the clock and the dog's brain goes somewhere else. I definitely need to improve my abilities at setting criteria and knowing when to raise them in the way that is most effective for moving the dog's learning along smoothly. Ideally the dog should swallow the treat and then go right back to doing the behavior, whatever it is -- a useful guideline to follow.
There was a brief discussion of types of reinforcers. GDB uses the dog's food for the most part as the reinforcer, which in most cases works well because most of their dogs are Labs. If they do identify a dog that is not food-motivated, they have a protocol for it, with the end goal of creating a dog that is food-motivated for its own food. At Leader we use Milkbone Minis for reinforcers, and I wonder how and why that got started. Was it because someone thought something different would be more reinforcing? That is totally possible, and I could see thinking that myself, but I wonder if it is really true. It seems like it would be a lot simpler just to use the dog's food. That way, too, if you pre-measure the dog's food you could just take out a handful of pieces for treats and know that the dog wasn't getting any extra food and there was no chance of it gaining too much weight.
Someone asked about play as a reinforcer, and Michele said it is not usually effective because it takes too long. Not only do you have to spend several seconds actually playing with the dog in order to make it reinforcing, you also then have to bring the dog "down" from play mode into work mode, which takes even more time. This is what I have always thought about play as a reinforcer for guide dogs, and I am glad to hear my opinion validated by an expert.
Michele also spent some time discussing the fact that it is not possible to have a purely positive training program. It is absolutely not true that positive trainers are permissive or that leash corrections can never be used. Undesirable behavior should be managed whenever possible, and when a leash correction does have to be used, the focus should always be on returning to positive methods as quickly as possible. So, if I understand it correctly, if my dog lunges to jump on someone, it's okay to leash correct him. But what it shows me is that my dog is lacking in self-control around people, and that's something I need to be working on before putting the dog in a situation where he needs to exercise self-control again. Also, when punishment has to be used, it should be negative punishment -- removing the opportunity for reinforcement -- as often as possible, rather than positive punishment such as a leash correction. Punishment can be used but should always be used in a thoughtful way.
After this general discussion of operant conditioning and training in general, we moved on to training specific skills. Pretty much every skill was presented like this: first a discussion and Power Point presentation, along with video clips for demonstration; second, a live demonstration from one of the conference faculty (in addition to Michele: Todd, Lori, Candace, and Jessica); third, hands-on practice with GDB dogs in training for the conference attendees who had working spots. I should clarify that when registering for this conference, we had the choice of an auditing spot or a working spot. Working spots cost more, but that is not the reason I picked an auditing spot. There were two reasons why I picked an auditing spot: one is that, in the presence of an expert, I freeze and wait to be told what to do, and two is that I honestly feel that intellectually I learn more from watching other people than I would if I were standing up there being told what to do. If I were brand-new with the clicker I probably would have paid for a working spot. But, while I wouldn't call myself proficient, I definitely know how to use a clicker. I was very happy with my auditing spot and would do the same again.
The first skill we practiced was introducing treat-taking and assessing whether the dog had any issues with treat-taking and whether it was or was not food-motivated. Treat-taking was first assessed just by feeding the dog kibbles and watching how it took them. This was done without the clicker. Then the Doggie Zen game was introduced. This is the same as the Doggie Zen popular in the pet dog training world: present the treat in a closed fist, which the dog will most likely lick and paw and possibly nibble at, and release the treat as soon as the dog decides to be polite and back off the hand. Once the dog understands this part, you bring your closed fist toward the dog slowly but stop out of range of the dog. If the dog stays sitting where it is, you give the dog the treat. If it lunges for the treat, you withdraw your hand dramatically and wait several seconds to reset. The goal of the exercise is to communicate to the dog that forward movement = sitting still and lunging for the treat = disappearance of the treat. The last part of this session is loading the clicker. The goal of loading the clicker is to make the dog realize that click = treat. You should try to click different behaviors each time, and not click repetitions of the dog staring at your hand. You should not do any more reps than necessary to make the dog realize that click predicts treat. Michele actually said you could skip this step, and move straight to training, because the dog will realize that click = treat very quickly. I'm glad she said this, because I gave up loading the clicker a couple of strings ago and found that it affected my dogs' ability to learn not at all.
The next skill was food refusal. You begin this by holding the food just out of the dog's reach in your right hand while holding the dog back with your left hand on the leash. As soon as the dog releases tension, you click and treat. (You can also do this with the dog tethered to something else, like a fence, to make it easier on your body. "Tether" is a word that we are hearing a lot here. It just means that the dog's leash is attached to something -- your hand, or something else solid -- or else you can use a "ground tether", which is stepping on the leash on the ground.) Once the dog shows that it understands the first step, you move on to the second step, which is dropping the food on the floor while keeping a short enough leash that the dog is never able to actually get to the food by lunging and thus reinforce the lunging. For the third step, you need an assistant. The assistant is at your left side offering food, and the assistant controls the food and does not let the dog get it. For the fourth step, the assistant sets the food on the floor next to the dog and covers the food with his or her hand if the dog tries to get it. For the fifth step, you walk the dog around the food on the floor while the assistant waits, prepared to cover the food if the dog lunges for it. (Walking around the food is an increase in difficulty from sitting still next to it.) Once the dog is showing that it understands that avoidance of food on ground = good chance of reinforcement from handler, it is then the handler's job to do regular set-ups in different environments to allow the dog to generalize. Food refusal is practiced every training session for the first week, and after that the number of deliberate set-ups is based on the individual dog -- its scavenging history, level of interest in food, et cetera. In advanced training, food set-ups should be made to look as natural as possible.
The obvious question about this food refusal exercise is this: I understand how the dog learns that it's never going to get reinforced by going for food on the floor, but that there is a good chance that passing up food on the floor = treat from instructor. But what is going to happen in the real world when the working guide dog passes by food on the floor, thinks about going for it, decides not to, looks up at its handler for a treat, and the handler doesn't give it one? Isn't the behavior of passing up food on the floor going to die out if it isn't reinforced heavily (since we are dealing with a hardwired instinctive behavior -- scavenging)? Michele says no, as long as the handler keeps up enough reinforcement for other things like curbs and targeting. When this behavior of food refusal is first taught, it does need heavy reinforcement -- much heavier than I have ever done in the past when attempting food refusal with my dogs -- but once the behavior is learned, it is easier to maintain. I freely admit that the skeptic in me still has a hard time buying it. But then I also believe that with a strong-enough reinforcement history -- which it is the trainer's job to build and, really, the client's job to maintain -- it is possible. Anyway, it's not like traditional methods did that great at controlling scavenging either. I have written a whole blog post on scavenging and I still pretty much stick by what I have written there.
I think every instructor or almost every instructor does some sort of food refusal training, even if it is as bare-bones as throwing a piece of food on the floor and correcting the dog if it goes for it. The difference here is the structure and the planned progression. For someone like myself who is a structured thinker and aspires to be a structured trainer, this structure is beautiful! Food refusal is not currently one of the things Leader uses a clicker to train, but we can always use a verbal marker instead of a clicker. Maybe not quite as strong, but the dog will still get the point.
After a tour of the campus, we get lunch, and the afternoon is all about the platform. The platform, in case anyone reading this doesn't know, is like a raised plank, about one foot wide by three feet long and maybe four inches high (although they can be made in any size or shape). It is widely used in competition obedience to teach proper positioning -- dog next to handler -- and GDB uses it for the same purpose. Dog-out-of-position is a big problem that comes along with the use of food rewards in guide dog training. Because you keep the treat bag on your right side (usually) and reach for the treat with your right hand, the dog anticipates the treat coming from over there and wants to curl around in front of you out of anticipation. This will lead to the dog getting out of guiding position unless you take steps to address it. One step is making sure to always deliver the treat directly to the dog's head at your left leg (or, sometimes, to the left of the dog's head if the dog is one that habitually swings its butt out to the left -- if you deliver the treat in a way that forces the dog to turn its head to the left, the hindquarters will generally swing to the right too). If the dog's head is reaching to the right, it doesn't get the treat; it only gets it if it is in position at your left leg. Correct food delivery goes a long way toward fixing this problem, but use of the platform can help as well. Generally, dogs like the platform. (I have dabbled with it with my current group of dogs and found that to be true -- they like getting up on it and perching there.) It makes it much easier to teach them to stand still at your left side because when they're on a platform, their feet are anchored there.
At GDB, they define correct position as feet facing the line of travel, body not angled, staying in that position while guiding and while waiting for reward. The dog may look in the direction the treat is coming from, but not move its feet. They use the platform to teach the dog that its feet must remain still, starting with the handler standing next to the dog's shoulders and then progressing to the handler moving back to the dog's hindquarters. The platform keeps the dog in place and also makes the handler aware of what correct position is. (I can totally testify to the necessity of the second part of this. When I was playing with the platform in the first weeks of training my current dogs, I noticed that it was really, really hard for me to stand with my hips square next to my dog on the platform. My left foot wanted to point in to the dog so much that when I made a point of straightening it, it felt like an unnatural muscle movement. I realized I have been standing with my left foot pointed in to the dog for a long, long time, and am now aware of it so that I can work on fixing it.)
Another thing Michele said that I thought was cool was that you do not have to free shape the platform. You can shortcut by using a lure. "Pure" clicker trainers like to free shape everything so that the dog has to think as much as possible. I, on the other hand, want to know the shortcuts because we are working within a time frame that can't be altered. Glad to know a lure is okay.
Once the dog has the skill of being on the platform and likes being on it, you can use the platform in helping to teach impulse control -- not going for distractions. In the beginning, the distraction needs to be at a distance at which the dog is able to control itself. It needs many repetitions and successes with the distraction at a distance before the distraction can be moved closer. This is the most challenging part of training at Leader right now. Because of the kennel construction, there is a serious shortage of low-distraction training environments. You pretty much have a choice of the kennel run -- where dogs bark all the time but there are no other distractions -- or else the hallway, which is not wide enough to get far away from another dog and person who are passing. Any distraction-free environments are so far away that by the time you got the new dog over there, its training session would be almost up. That is just going to be an ongoing problem that we will be dealing with for another 14 months or so, I guess.
Another cool thing you can do with the platform, once the dog is happily getting up on it, is to add another small square platform at the end of the first long rectangular one. This small one is about twice the height of the original one and is meant to introduce the concept of keeping the dog's body straight while targeting up curbs and up stairs. Really the platforms look not only extremely useful, but also lots of fun. They come in different primary colors and sort of remind me of a big set of blocks for grown-ups.
After watching the platforms all afternoon, my brain was so full I could barely summon the energy to take notes. Hard to believe there would be four more days of this stuff. I just hoped my brain would be able to function again the next day.