Learning from (no) Mistakes

Errorless learning is a teaching procedure in which the learner does not make any mistakes, and doesn’t need to in order to learn.  

Trial and error learning is a teaching procedure in which mistakes are deliberately included to provide the learner with information. 

Most dog training procedures are based in trial and error learning.  Dog trainers set out to teach their dogs a task and they not only anticipate that the dog will sometimes get things wrong from the get-go, they feel that the withholding of reinforcement (or, if we are setting the clocks WAY back, the “correction”/aversive stimulus) that occurs in conjunction with the error is actually helping their dogs learn.  This is not often argued against in the dog training world.  In fact, most people involved in dog training will claim that without mistakes the dog never truly understands how to be right.  This is said across the board regardless of the training goal.

Go ahead, suggest to a group of dog trainers that we could be training dogs with errorless learning.  Have fun with that. I’m not responsible for what happens to you.

This video is me training my puppy an impulse control skill.  I am using a trial-and-error learning procedure that is common in the agility world. You will see that he is not traumatized and he learns quickly. This is what most dog trainers will argue when you suggest we could teach without errors.

But I am going to pick this video apart and talk about the data.  Felix doesn’t get any food until about 20 seconds into the session, so that was 20 seconds of errors before any success occurred.  In this roughly minute and a half training session Felix was given 15 bites of food; meaning he was successful 15 times. And how many errors occurred? I counted 23.  So in this session, 23 behaviors were errors and 15 were correct.  Not exactly a ratio I am in love with.

Conversely, let’s look at the same behavior taught using an errorless learning procedure:

In this video I teach Brink he will get food for staying away from the food by simply keeping it out of his reach at first and moving it marginally closer to his face on each repetition.  He makes zero errors.  He is fed at about 5 seconds into the session, and he received reinforcement 46 times (in almost 3 minutes, which means I am slow). He made what I would call a “pre-error” (he moved closer to the palm) 7 times.  Each time he was paid anyway (because he did not make the error, he was still right!) and I responded by moving my  hand marginally further from his face on the next repetition.

So, simply by numbers, the dog in the second video gets reinforced more often and makes far fewer errors than the dog in the first video.  If learner frustration is something you care about, that’s probably all you need to know.

“Errors are not necessary for learning to occur. Errors are not a function of learning or vice-versa, nor are they blamed on the learner. Errors are a function of poor analysis of behavior, a poorly designed shaping program, moving too fast from step to step in the program, and the lack of the prerequisite behavior necessary for success in the program.”~BF Skinner

Like I mentioned above, learner frustration is a real concern of any skilled trainer. A frustrated animal is more likely to act out aggressively and less likely to continue to participate in training in the future.  If we are to create engaged and willing partners in our dogs we would do well to avoid causing them frustration.

We can all agree that frustration feels yucky, and that getting stuff right feels good.  Never forget that classical conditioning is always occurring–do you want your dog to associate the yucky feeling of frustration or the great feeling of getting stuff right (and getting reinforced!) with your training?

The bottom line is that to design an errorless learning procedure for your training projects is to design a solid plan that will get the job done with minimal frustration (for both of you!).  Creating procedures that follow this principle can be a challenge at first, so here are some tips:

  • Make a plan, and set the scene for the behavior to be likely to occur. This might mean using a hallway to train your dog to walk backwards, or it could mean delivering reinforcement in such a way that the dog is set up to repeat the behavior.
  • Break the task down into very small manageable steps. If there are errors, you may have a problem here.  In the fist video, I am asking for the entire behavior right off the bat, and so a lot of errors occur.  In the second video I am only asking the dog to do tiny slices of that final behavior I am looking for.
  • Layer difficulties into the picture in such small doses they are not even noticed by the learner. In the second video if I moved my hand to close to Brink’s face he “noticed” the hand had moved, and I always had to back up.  Ideally this never happens, but if it does, it’s best to catch the error before it occurs, and back off the difficulty for the next repetition.

In this video, I am introducing Felix to the directional cues I will use for him in agility. I have designed this procedure with errorless learning in mind. I set him up for success in every step, and when I introduce difficulty (a competing toy) I not only use a lower value toy, but I place it VERY far away, almost ensuring that he gets the answer right.

 

So, dog trainers, I challenge you. Take a trial and error procedure you’ve been using and rewrite it be errorless. You’ll be amazed at the results.


20 thoughts on “Learning from (no) Mistakes

  1. Really great post! I was wondering if you could explain the relationship between impulse control and frustration tolerance with respect to errorless learning? By going through an errorless learning procedure are we teaching frustration tolerance? Is that even a ‘thing’ we care about anymore if we follow errorless learning procedures?

    1. You’re thinking really smart! I think you hit the nail on the head with your final question about whether that’s a thing we should care about. As I mentioned in another comment, my next post will concern frustration tolerance in animal training. Thank you!

  2. Could you please comment on how errorless learning fits in with building resiliency and ability to tolerate frustration/error? In real life, we all make mistakes, especially during a sports performance. While I see benefits of errorless learning, I’m not sure what the impact is when actual mistakes happen. (Since I compete in agility, I’m particularly interested in how this learning might fit in.)

    Thanks for the the thought provoking blog!

    1. Excellent question. I do believe the trial and error version of itsyerchoice is *designed* to build the type of resiliency you seek. I believe their are better ways to help dogs build up life-required frustration tolerance. I personally don’t think they should have to tolerate frustration when it comes to our training; I put it on me to be sure I have designed a plan that does not create frustration. I believe you just inspired my next post with your question, because I am finding myself wanting to write several paragraphs here. Thank you!

    1. Hi Kathy, I did my best, and all I can give you is the date of the paper, not the paper itself, which is 1968. Here is a true citation of the same concept: Terrace, H.S. (1963). Discrimination learning with and without “errors”. Journal of the Experimental Analysis of Behavior, 6, 1-27.

  3. I’ve heard many people talk of this for basics…sit, down etc. But my brain just doesn’t see how. And it appears to me that you can’t do true shaping and be errorless at the same time.

  4. Why not reward further from the hand as the desired behavior is to stay Away from the food. Rewarding so close to the food is confusing to the dog…rewarding farther away makes it more clear.

  5. I couldn’t agree more. I always set up dogs I train to succeed. Think it out carefully first so that your dog is successful every time. Success breeds success; no truer statement.

  6. In the first 2 videos, the trainer is making many mistakes with itsyerchoice, therefore it looks like the dog is making several errors. I have used this method on my puppies at 12 weeks old and they make less than 5 mistakes before getting it right all the time. In the 2nd video, I see more errors by the dog than in the 1st. It also comes down to what is reinforcing go the dog. Getting to smell the treats in the hand during itsyerchoice is technically a reward, according to the creator of the game. Not disagreeing with error vs. errorless, but not seeing a difference between the first 2 videos.

    1. How do you keep a dog from smelling food? What errors is the dog making in the second video? He is not being asked for a specific behavior, just not to grab at the food, which he only moves forward a couple times. And no, that dog was never trained itsyerchoice.

  7. This is a new concept to me entirely. I often feel frustrated and so does the dog with the other method, but being new to dog training I have a specific question: how would you, say, train the dog to stay in a specific spot and not rush the door as soon as you walk in that direction? You have sit, stay, and wait-until-I call-you. But to combine all three seems overwhelming.

    1. Here is one paper:
      Terrace, H.S. (1963). Discrimination learning with and without “errors”. Journal of the Experimental Analysis of Behavior, 6, 1-27.
      And here are several things you could look up yourself if you care to:
      Skinner 1968
      Chance 2003
      Pierce and Cheney 2008
      Thorndike 1898
      Rosalez-Ruiz 2007

  8. Reblogged this on Historien om Alf and commented:
    Mycket intressant blogginlägg om hur man tränar med belöningsbaserad metod istället för “trail and error”. Ska verkligen tänka mer på detta 🙂

Comments are closed.