This is a good example of one of the many handouts we provide to our DogNostics Dog Training and Behavior students as they work through our programs. This handout was prepared by Angelica Steinker. DogNostics is the educational arm of the DogSmith and is also open to the public for a selection of training and behavior program.
Mechanics: Training with the use of a “reset cookie”.
Reset Cookie Definition:
A process that makes use of giving a free cookie or reinforcement that is non-contingent of the goal behavior to avoid loss of motivation. If you lose a learners motivation you have nothing to train, thus the use of a reset cookie while undermining stimulus control is valuable to the training process and likely falls within the realm of the art of training.
Some extremely successful trainers will argue against the reset cookie. As a result Angelica contacted an expert in respondent conditioning.
Question: On 10/12/2013 11:34 AM, Angelica Steinker wrote:
Hi Professor Rescorla,
I am a dog behavior consultant and a faculty member at two online dog trainer trade schools. I am looking for clarification on a training question.
While shaping when the trainer makes a mistake (missing the opportunity to click) or needing to restart the dog from a specific angle, we often use something we call a reset cookie.
We use this reset cookie in the following way, we do not click and simply toss the treat (for free) to a specific area. We do this to maintain a high rate of reinforcement, avoid frustration, and as I said above to get a specific approach angle to the shaping area.
At a seminar with a very high level behaviorist we (the attending people) were told to never use a reset cookie. I am concluding this may have been a comment not referring to shaping but rather to training in general. However I also wonder how the use of a reset cookie can be bad training. If you consider this example:
Trial 1: dog fails
Trial 2; dog fails
Trainer moves dog to another area, gives a free cookie as a jump start for motivation, and then making adjustments goes for a third trial (which arguably is really a new trial as you paused and moved to another area.
You were cited as the person who has done research that clearly shows that using any reset cookie is damaging to the learning process.Could you kindly clarify to me if my thinking is off and if so what study I may read to deepen my understanding?
Answer from Professor Rescorla [cited with permission]:
In Pavlovian conditioning (respondent conditioning), in which animals are learning the relation between a signal and consequence, it clearly hurts the learning if the consequence occurs at times other than following that signal. In effect, the animal is asking how good a predictor is the signal; it is an inferior predictor if the consequence comes even when the signal is not given.
In instrumental learning (operant conditioning), which you are doing, life is more complicated. To be sure, giving the food at a time when the animal has failed to engage in the target behavior rewards it for doing other things. That will slow learning. But it is also important to maintain the animal's engagement and giving a "free" reward can sometimes be important in maintaining motivation. If you were, say, to give the animal the reward whether or not it does the required behavior, it would never learn. On the other hand, if the animal quits on you, he will not ever engage in the target behavior, so he cannot learn the task.
One of the things that makes shaping an art is sensing the tradeoff between keeping the animal engaged and occasionally rewarding him for incorrect behavior. If you engage in a training procedure in which behavior X only rarely occurs, it is common to first reward behavior as much like X as possible and then gradually increase your demands for how like X must be the behavior to earn reward. As you do this, you will probably have to occasionally reward behaviors that reflect backsliding away from X in order to keep the animal alert and active. Doing so also rewards competing behaviors, but may in the end facilitate the whole process.
I hope these comments help.
[Cited with permission]
Angelica Steinker DogNostics Faculty 2015