Training 102

Training 102

Training, as with any other thing, works well when done correctly. It’s hard to train an animal if you aren’t consistent, systematic, and variable. A well-known marine mammal trainer, Karen Pryor, has a list of rules to ensure success in training a behavior.
1) train the behavior in small enough increments
2) train one aspect of the behavior at a time (even a jump has several aspects, such as breaching the water, arching the body, etc.)
3) relax other criteria when introducing a new criterion (criteria are the standards by which something can be decided, such as whether or not a behavior was performed correctly)
4) don’t shape for multiple criteria at one time
5) stay ahead of your subject—what if you’ve only planned three steps in training a behavior and suddenly your dolphin predicted what to do up to step six?
6) don’t change trainers midstream
7) there are as many ways to shape a behavior as there are trainers to think these ways up
8) don’t interrupt a session gratuitously
9) quickly review the whole shaping process with a series of easily earned reinforcers when coming back to a training session
10) end each session on a high point, where the animal has felt (as have you) that it has succeeded

The basic synopsis of good training is simply, be consistent, stay on your toes, make it worthwhile for the animal and be precise and complete in what you’re asking of the animal. This process, called shaping, or selectively reinforcing a desired response, is not perfect. The animals are not perfect, and neither are the trainers. Trainers, in fact, often engage in superstitious behaviors that may actually hinder the animal from properly learning a behavior. A superstitious behavior is performed under the notion that it will render a specific response. More often than not, trainers don’t notice that they do it. The problem with this lies mostly in crossing over between trainers. What happens when a dolphin who is used to taking an Sd from a trainer who always puts her hands on her hips before giving the signal gets asked to perform the behavior by another trainer? That dolphin may not perform the behavior because the hands on the hips have become a part of the signal. Dolphins can also have superstitious behaviors. For example, one day a very good trainer went out to give a show at a marine mammal facility. He’d been doing the show for years and years and suddenly the dolphins wouldn’t even come to the station to begin the show. Finally they brought out another trainer and the show began. The next show he tried again, and they still wouldn’t come. Turns out, the trainer had bought new boots. His old boots had had black soles like most shoes, which the dolphins saw before each show at the station. His new boots, however, had white soles. Those dolphins only came to the platform to black-soled shoes.
Training can also often involve encountering unwanted behaviors. For example, in the 80’s at Dolphin Plus, all of the trainers were male. They trained the dolphins to pull at a string when presented with one, which they then translated to taking off women’s bikinis. The male trainers thought this was great, so they reinforced it. Then, along came a female trainer who was having none of it. They stopped reinforcing the behavior, which only caused the dolphins to do it more. This is called an extinction burst, where a previously reinforced behavior suddenly stops receiving reinforcement. This causes the animals to repeat the behavior over and over again until they finally give it up, which is called extinction. A great human example of this is with elevators. Say you live on the 15th floor of a building and every day you take the elevator. One day, the 15th floor button won’t work. You push it again. And again, and again, and again. Eventually you give up and offer another behavior, walking up the stairs, or taking the elevator to the 14th floor if that’s an option. Chances are, you’re going to try the 15th floor button again the next day, and the next, until you finally give up completely and revert to taking the stairs every day. Extinction.
Another way to stop a behavior without giving punishment is by simply ignoring it. When an animal is asked for a behavior and gives another, the trainer will give an LRS, or least reinforcing scenario, to that animal. This involves a short, 3 second “step back” from the failure, which gives the animal a moment to regroup its thoughts. Simply asking for the behavior again right after failure can be confusing, and so the pause is used to tell the animal, “No, that wasn’t it, now calm down and we can try again.” The LRS can be used to reinforce calm behavior, which helps with increasing the chance of success after an initial failure.
You’d think that training animals to do unpleasant things, such as give blood, allow a tube down their throat for hydration, etc., would be difficult. In fact, it can be, but the use of positive association is a powerful thing. The Premack Principle says that a high-probability behavior reinforces a low-probability behavior. That is, if you prick a dolphin with a needle (low probability) and then let it go do its favorite behavior (high probability), it’s likely that they’ll begin to acquire a positive association with the once scary behavior.
Do you ever wonder how someone could possible train a dolphin to do aerial tricks with a simple clap of the hand? Do you just sit there clapping your hands until they do the right behavior? Of course not. A practice called fading is used to gradually change the stimulus controlling an animal’s performance to something more reasonable, like a hand signal. Typically, the first stimulus used is a target pole, a prop that pinpoints a specific location for an animal to touch. By using a target pole, you can teach a dolphin to jump up, because it has to jump up in order to reach the target. Eventually you can fade that stimulus by introducing a hand signal when the behavior is learned, and then eventually completely remove the initial stimulus from the equation. Voila, back flips for petty golf claps.
Some behaviors simply can’t be taught. While you can ask a dolphin to manipulate its movements by having it target a pole, you can’t force it to do things like vocalize. All vocalizations have to be captured. When a behavior, like a click, is spontaneously offered, it can be reinforced and paired with a stimulus that will eventually elicit the behavior. The problem with this type of training is that there are no actual steps, which means that if the behavior suddenly disappears, there’s no way of backtracking to retrain it unless the animal offers it again.