Science Sunday posts are short posts about the science of animal behavior and training. They often feature a quote or a passage of text. Spend a moment today thinking about the ideas in the post. As always, you can share your thoughts or questions in the comments section.
“A program is a defined set of procedures by which behavior, behavior-contingencies relations, and the contingencies themselves are changed. There are different ways to program. One way is through trial and error. However, in this kind of learning the individual makes many errors and can suffer in the process of learning. He can also learn through a trial-and-success program. This is designed in steps which are attuned to, and built upon, his progress. In the course of this kind of learning he will not make mistakes. Both methods of learning may produce the same pattern, but, although the outcomes of learning are the same, there may be important differences in other areas. For example, the child who is taught to swim by being thrown into water over his head may eventually swim as well as the child who is gradually introduced to water. However, the first child, once he is away from his watching parents, may shun water. One can say that they have had different histories or experiences, or one may state that each has been involved in a different program.”
Arthur Schwartz & Israel Goldiamond
Social casework: A behavioral approach, 1975
Learning usually isn’t much fun when you just can’t figure out the right answer. Think back to a class you took in high school or college that was impossibly difficult. Perhaps it was high school trigonometry, a college Spanish literature class (taught in Spanish!), or a world history class that involved memorizing hundreds of facts. Were you relieved when the course was finally over?
On the other hand, when a person (or animal) is met with success during learning, the person usually has better retention of the information or skill, enjoys the learning process, and looks forward to learning more. In addition, the teacher is usually regarded more favorably.
Trainers who train with shaping and positive reinforcement are often able to train behaviors in such a way that the animal makes very few errors during the acquisition of a new behavior. The animal is successful during learning because the trainer breaks the task into logical steps and sets up the teaching environment so that the correct behavior is very likely to occur at each step. Here’s an example of a creative trial and success program for teaching a dog to go to a mat.
Watch on YouTube: Target training to the mat
Of course, as Goldiamond and Schwartz discuss, there are usually lots of different ways to teach a person or animal to perform a behavior that result in similar outcomes. However, even when the outcome is similar (for example, swim two laps in the pool), an observant individual can often notice differences in performance if learners were taught with different programs. This could include differences in body language, as well as the person’s willingness to engage in the task.
If you want eager and engaged animals that are willing to learn and perform, think about how you can make your training trial and success learning, rather than trial and error.
There are those (Bob Bailey and Ian Dunbar among them I think) who say that mistakes (errors) are an essential part of learning. I think it’s an unavoidable part actually, although I try to aim for what you describe and quote above. Do you have any thoughts on this? Is there an ideal ratio of errors to successes? Do we need errors to build resistance to inevitable mistakes and persistence in learning?
it probably depends a great deal on how you define a ‘mistake’. It can be ‘doing the wrong thing’ or simply ‘not doing the right thing’. With ‘positive’ training we tend to not think of ‘not doing the right thing’ as a ‘behaviour’, so therefore it is not a ‘mistake’. We tend to wait it out, recue or reconsider out training method and criteria.
Then different individuals have different learning styles and personalities. I have always hated ‘trial and error’ learning — I thought that it was just mean of the teacher, who KNEW the correct answer, to not simply tell me. I did enjoy ‘practical’ work where we demonstrated to ourselves how things worked 🙂
I think dogs can be like this too. “Shaping” with the German Shepherds and Kelpies can be a thankless task. They say, “Just TELL me and I’ll do it!” But I’ve now got a little dog (Beagle/Cocker Spaniel) who thinks ‘shaping’ is the best game ever. Her attitude seems to be that if I tell her what to do, I have some nefarious purpose and she spooks. She has no worries about ‘mistakes’, — if that doesn’t bring the treat, then she’ll try something else.
The ever present “it depends” 😉
Hi Jenny,
You raise some really great questions and points!
I think a lot of the research on errorless learning has been done with tasks that are fairly “black and white” — everything from teaching a pigeon the difference between red and green to teaching neuroanatomy terms and concepts to medical students.
This gets interesting when we start talking about shaping, when the animal is potentially offering lots of different behaviors, some of which we want, some of which we don’t want, some of which are approximations to what we want, and some of which are previous approximations which we wanted yesterday but don’t want any more.
As with your dogs, I think the learner’s history (and probably also breed / genetics) can play a role in whether the learner will do better with more guided learning, and how much guessing / experimenting should be included.
This can also change depending on the task / findal behavior. (As a silly example, imagine using free shaping where the learner does a lot of guessing to teach someone who has no experience with power tools how to use a chain saw to chop down a tree. Even if you’re clicking the correct approximations, they might lose an arm in the process!)
Cheers,
Mary
Hi Jane,
Great questions and points.
There’s been some really interesting laboratory research on errorless learning. For instance, in the 1960s, Terrace trained pigeons to peck when a vertical line was displayed and to wait and not peck when a horizontal line was displayed. This is something that pigeons can be trained to do, but that often involves a long process with many, many errors (pecks to the incorrect line). Terrace was able to do it with zero errors. He measured other behaviors too, including “emotional” behaviors (wing flapping, stomping, etc.). Birds who learned without errors were quite and calm, compared to the quite frustrated birds who had to figure out the task by trial and error.
Interestingly, one of Terrace’s other experiments showed that learning something with a lot of guessing and errors can then disrupt an earlier behavior that was learned without errors. (Basically — the birds learned a first task without errors, then a second task with many errors. When the researchers then repeated the tests from the first task, the birds made some errors.) In short — errors led to more errors.
I would disagree with Bob Bailey and Ian Dunbar — I think there’s enough research to indicate that mistakes don’t have to be part of teaching. Taking this research and actually implementing it when training is something that I think we’re still figuring out and probably will never completely figure out — as there can be so many more variables and complexities when working in actual training situations.
So, in the “real world” it is usually difficult, if not impossible to avoid all errors. But, I think it’s worth thinking about when designing training programs. As well, I think that tasks that result in a lot of errors during teaching or that we label as “hard” or “difficult” to teach indicate that we have not yet identified all of the component skills, pre-requisites, and most effective teaching methods.
Interestingly, I think we are actually getting quite good at teaching animals with positive reinforcement and minimal errors. However, where I see a lot of gaps currently is still in how we teach people to train animals. I think this is why quite a few people give up and go back to their traditional methods — they give clicker training a try and find the learning curve of figuring it out too confusing and frustrating because they are making mistakes and unsure of what to do.
Best,
Mary
Thanks Mary! I like your answer (meaning I prefer to train in such a way that minimizes errors even if I can’t completely eliminate them). And yes, teaching people to train is a more complicated process. One problem is that we can’t control the environment of those people. I think if I could be present for each and every interaction of the handler and animal, it would be different, such as each time I bring an animal to a training session. But these people want their own lives 😉
I searched for Terrace (and got lots of lovely photos of pigeons on terraces 🙂
But found this:
http://onlinelibrary.wiley.com/doi/10.1901/jeab.1963.6-1/abstract
and more.
It seems that he is describing what we, in the dog world, call
“breaking it down”. Making it almost impossible or the learner initially to get it ‘wrong’. I like that! 🙂
But yes, some people (of any species) seem to enjoy a challenge. (My favourite challenge is cryptic crosswords — too easy is boring, too difficult is frustrating.) I think that the challenge of a teacher is to find just the right degree of challenge for EACH learner.
As a classroom teacher I found THIS frustrating, especially in non-graded classes 🙁 But when training animals I find a joy in finding the right balance to maintain my ‘student’s’ interest 🙂
Hi Jenny,
Yes, the paper in that link is one of Terrace’s more famous studies.
His work involved arranging the environment and gradually shaping the behavior so that the pigeon (or other animal) was successful nearly all of the time.
People usually think of errorless learning as small steps. However, one thing that Terrace did and which helped him reduce errors during teaching was that he would often start at a place where the animal could already be successful.
So, he would start with skills or behaviors that the animal already knew, and then shape toward the new behavior. I think good dog trainers do this all the time, and if we stop to think about this, it can really help our training. Rather than thinking about starting from zero with nothing, think instead about what behaviors / skills the animal already knows that we can begin with at the beginning of a training program.
cheers,
Mary