This post is a continuation of my notes from the 2017 ClickerExpo conferences. This year I had the pleasure of attending both the January and March Expos because Dr. Jesús Rosales-Ruiz and I taught two lab sessions involved the shaping game PORTL. You can find more of my ClickerExpo notes here.
How does shaping work?
The funnel metaphor
At the January expo in Portland, I attended a lecture by Alexandra Kurland about a concept she calls “loopy training.” If you’re not familiar with this term, I suggest visiting this blog post, where I explain more about the concept of loopy training and discuss why thinking about loops can help make you a better trainer. (If you really want to know more about this topic, you should also check out the five-hour DVD that Alexandra has on the subject of loopy training.)
During the lecture Alexandra talked about funnels. When teaching the concept of shaping, trainers often use the analogy of a funnel to explain how shaping works.
Think about a funnel. It begins very broad on one end and then gradually narrows down to a tiny opening at the other end. Likewise, this is how many people approach shaping. The trainer begins by reinforcing very broad approximations. Anything that is even somewhat close to the target behavior will earn reinforcement. The trainer might sometimes even reinforce a few behaviors that don’t have anything to do with the final behavior, if she feels the animal needs a bit of extra encouragement to get started.
Then, the trainer begins to gradually tighten her criteria. Now, only certain behaviors earn reinforcement. Over time, she reinforces behaviors that are closer and closer and closer to the goal, until the animal is now performing the final behavior.
If you’ve ever tried shaping a behavior this way, you know that it can be somewhat of a messy process at times. During the initial stages, the animal offers lots and lots of behaviors. Many different behaviors get reinforced as you help guide the animal to the final behavior.
All of your intermediate approximations become part of the animal’s history. This can help you later on because if the animal is having trouble, you can return to a previous step.
However, all of these different versions of the behavior that are reinforced during shaping can sometimes also get in the way. If the animal is distracted, disrupted, or stressed, he may revert back to earlier (unwanted) variations of the behavior that were reinforced during the training process.
It is also important to note that depending on the experience level of the animal, the experience level of the trainer, and their relationship, shaping this way can be a lot of fun, or it can also be pretty frustrating for the animal and/or the trainer.
Flipping the funnel upside down
During the lecture, Alexandra also discussed a second way to think about shaping. Instead of beginning at the top of the funnel with broad approximations, you can flip the funnel upside down and begin at the narrow end of the funnel.
What would this look like?
The trainer could start with a very small, clean piece of behavior in the beginning. This may (or may not) be a “simple” behavior, but it should certainly be something that the animal can already do successfully, with no (or very minimal) errors.
The animal already knows what to do, rather than guessing and offering lots of different behaviors. As a result, there is not a great deal of variability in the animal’s behavior, even from the beginning.
Then through a series of carefully-guided steps, the trainer builds on and expands this behavior. The behavior shifts slowly into a new behavior. It may become increasingly longer or more complex until the goal behavior is reached. This type of behavior may be stronger if challenged, as the animal did not rehearsed errors or practice lots of unwanted versions of the behavior during the shaping process.
Alexandra discussed that both approaches can be useful for training. I think which approach is best can depend on a variety of factors, including the particular behavior being shaped, how the behavior will be used later on, the animal’s history, the trainer’s skill level, and other variables.
Alexandra gave an example in her talk of some training she has done recently with a friend’s Icelandic horse. During each session, she begins with some carefully-designed in-hand exercises. She picked a starting point initially that she knew would result in a clean, tight loop. This is the upside down funnel approach.
During the second half of the session, her approach is more loose. It is the usual, right-side-up funnel. Originally, she would walk around the arena and provide reinforcement if the horse was near her. This evolved into walking together and then into an exercise Alexandra calls “magic hands” and more recently into some lovely lateral work, which is related to the in-hand exercises from the beginning of the session.
The two funnels now converge in the middle. However, Alexandra did say that the lateral work during the second half probably would not have emerged without the other in-hand exercises as a foundation.
How can you apply this to your training?
This week, I’ve been thinking about some of the things I’ve trained recently and which funnel best characterizes my approach in each situation. I encourage you to take some time to do the same.
Pick a behavior you’ve trained recently or that you plan to train soon. Do you start broad and reinforce anything that is close to your goal? Or begin very narrow? Did you deliberately pick that approach or is it just the way you usually train that behavior?
Does your shaping generally take more of a right-side-up or upside-down funnel approach?