It's a common idea: 100% reward schedule is the most effective. It makes sense to a human brain. If I know I am always going to get something versus taking a chance that I might not? Which will make me respond more quickly?
It's a typical and understandable human viewpoint, since our brains look for patterns, not probabilities. Dog (and most animal) brains use probabilities. On a multiple choice test, a human would look for a pattern, such as the correct answer to the question! A dog would pick whatever was the most commonly used right answer on the last test, say, answer B, and answer all the questions with that answer. (Actually, I do suspect that some dogs have pattern recognition capabilities, and I need to do research on this. Or someone does. Go for it, anyone who cares to "steal" this idea and run with it!)
The "variable reinforcement schedule", where only a certain percentage of correct responses are rewarded, in a random fashion, has been supported as most effective via thorough research. "Always", 100% of correct responses get rewarded, is a CONSTANT reinforcement schedule, and less effective.
Why is it most effective? Well, here's where I am out on a limb. *crossing my fingers for acceptance to grad school -- Let me in, people, I got work to do!* Research would be necessary to say that this is fact.
A variable reinforcement schedule (NOT getting something every single time) may activate the emotion that Temple Grandin refers to as "seeking". "Seeking" is one of the primary emotions she identifies, and dopamine is involved. Dopamine is the brain chemical responsible for making things "just feel so right" -- some might call dopamine a reward your brain gives itself. By withholding reward on some trials, we enable the activity to become a seeking activity, and therefore an even more rewarding activity.
The question is, what level of reward, percentage-wise, will activate seeking WITHOUT activating frustration? Again, research is necessary, but I'd expect to find it varies by dog, although there may be commonality among groups -- breed, behavior tendencies, "IQ", or even personality type.
What is your experience with your dog's recall? Have you had different recall experiences with different dogs? What about with different training approaches?
Subscribe to:
Post Comments (Atom)
I find that Mungus' recall trend corresponds to the inverse of the number of visible squirrels, something like (Recall)= 1 - (# of Squirrels)
ReplyDeleteWho is Mungus? I thought it was Max.
ReplyDeleteWhat would you say is the reward schedule presented by the squirrels? Does the dog get rewarded by the squirrel hunt EVERY time? Is a "reward" highly probable? Or just highly desirable? Is it the squirrel-SEEKING behavior that accounts for the incredibly reliable attention the squirrels are able to get from the dog?
I don't have an answer for this. Just wondering about it. Be the squirrel?
Max's recall has been very, very poor for the past few years. Mungus generally comes when I call him now.
ReplyDeleteThe squirrel is its own reward, so I would say that the reward is highly probable. Who else plays with you every time they see you?
Right. But his stay is unbelievable.
ReplyDeleteSo, play EVERY time is the reward? Is play such a powerful motivator that even using what is typically a less effective reward schedule (100%) EQUALLY as effective as the typically more effective variable schedule? Or is there a component of variability that we are not recognizing?