The Need to Correct an Error
Our cultural history of training dogs is laced with Person Shows Dog types of principles. As if a person can tell a dog how to sit when every dog knows how to sit at 3 weeks old. The training process focussed on removing what was not desired and leaving the residue as correct. Sometimes this was identified with praise.
How we subtract that “not desired” is the biggest question today.
The same mindset is still prevalent in many training models, even those centred on positive reinforcement. The programs seeking immediate obedience are centred on the error, or slow response, which is punished through verbal or physical prompting. Some are less kindly than others and subscribe to a belief that unless the dog knows what is wrong they cannot be selecting what is correct.
When learning is littered with costly experiences of wrong even if we are doing well we will carry the anxiety of avoidance. Doing well becomes tainted.
You Are Wrong
To go wrong carries substantial emotional avoidance:
Are we stupid for not knowing better?
Is it our fault that we did not remember what to do?
If we cannot understand is it our burden to work it out, whatever that skill is?
Making a mistake, and not having sufficient knowledge or the skills to “think about it” are threaded through much of our life so we develop good skills at hiding error or simply not trying in fear of going wrong.
If you have stepped into the Apple Learning Program you will need to have no fear of error otherwise you will not learn. Exploring and experimenting: “oh, I didn’t know I could do that” …. is their learning model. No instruction book with Apples: switch on and have at it. If your computer learning predates Apple then you will be conditioned to be very afraid. Didn’t save it? Tough. Power cut? Tough. There was so much that could go wrong and valuable work be lost forever that hard and fast rules were obeyed. College programs for the 80s’s evening classes usually scheduled Computing for the Terrified.
Can you imagine the continual anxiety associated with a tool we now use every day. Do we want people to be fearful of using technology or eagerly seeking it? Do we want our dogs natural desire to learn to be self-rewarding and developed or perpetually anxious about making mistakes?
We should protect our young learners from harmful errors, but neither should they be afraid to experience error and learn how to assess what happened and make the changes. That is a skill to teach and ensure is in place before the mistakes occur or are beyond the learner’s skills to resolve.
When we watch puppies learn, errors are feedback: they will occur when the outcome the pup is seeking does not happen. Run too fast on a slippery floor and you crash into furniture, run too fast to grab a toy and you overshoot and need to turn back. It is the dog that gets to decide whether the cost is worth attending to and changing behaviour next time around. But do they have the knowledge or understanding of what skill is missing or how to learn it? Sometimes these error outcomes are of little consequence to the dog but a significant consequence to us. A dog crashing into people can be costly.
Error Learning is a Part of the Plan
Protecting the learning from error is a decision: is the error effect likely to be of benefit to that learner or increase anxiety associated with that activity? Learning confidence and self-esteem are intricately linked and often associated with specific activities.
Learning develops through different stages: in the initial steps the protection from error should be 100% so that confidence and competency grow alongside each other. Once the competency is of a sufficient level the learner will be asked to apply what they have learned to different situations. This is where the error likelihood increases and gives information on what has been learned. The learning gaps.
Examples:
Pup is taught to sit, on the word “sit”. Pup is in front facing the person. All is good. This seems to work and is learned.
Does this learning apply to the pup at the person’s side? Does this occur if the person is facing away from the pup? Extremely unlikely. The learning is within very specific parameters which need conscious variations to secure the learning competency so that the pup can connect the conditions to other situations.
This is a shortfall in the learning plans. Not an error burden to be carried by the pup. It is sometimes called the “but he knows sit ….” excuse.
Micro Errors will Enhance the Learning
Knowing that we will want the verbal cue responses to occur in many situations we gradually teach those discrimination skills. We plan micro errors: instead of giving the cue “sit” we say “sandwich”. For the pup that does not respond, or gives the “huh?” expression we reward. “Sandwich” is a non-cue: therefore non-response is correct. Reward.
But should the pup respond because hearing “blah” is their learning to this point, we need a way to give feedback: “Sorry, no bananas, I cannot be giving you five-star goodness for that response”
We give feedback that the anticipated reward is no longer coming, but this response is NOT EVER something for the dog to be fearful of. No pops, slaps, jerks or physical correction.
Here we need to explore the Reward Prediction Error: the difference between what you expected and what you got. I will leave you to learn about RPE it is a mighty chewy topic. Without RPE learning would not be moving forwards. There needs to be a distinction between the best rewards and minimal rewards: and do be careful of silence as a no reward, it can be quite aversive. Your minimal rewards should be along the lines of a conversation of that being a very interesting response but my hand is not going into the pocket to collect food for you.
Our judgement will revolve around the reward value of the choice we have just made: there will be some reward, but as we all know some rewards are better than others.
Make sure the expected reward can step down but equally make sure the unexpected reward to that choice is not so extraordinary that the dog cannot reflect on what was the difference. That jackpot effect may not be your friend, but a lower reward is still a reward, not a zap.
Before you go to make a food pot of variable treats make sure your dog can discriminate between Good Treat, expected 🌟🌟🌟🌟🌟, Average Treat 🌟🌟, and yeah well … 🌟. If delivery of the one-star food does not elicit a behaviour change then it is not functioning as a lesser reward. This is assuming that the dog is quite aware and demonstrates what and how to do it for the Good Treat.
Delivering the one-star should immediately prompt a return to the level of five-star worthiness. If the pup does not know how to achieve five-star then they cannot create it.
I distinguish between:
🌟🌟🌟 = much-loved food (as is all food)
🌟 = sorry, no bananas, no food, but I still think you are adorable
🌟🌟🌟🌟🌟 = much loved food and serious adoration, but not so much you get to explode with enthusiasm (tailored for individual capabilities of fan appreciation)
Error reinforcement
Yes, this is a thing. For both people and dogs making mistakes, errors or incompetency can bring greater attention than getting it correct. We can inadvertently reward poor achievement or non-compliance. This is most likely to be seen in the recall where the person is so darn grateful that the dog returned at all that they always pay out whether the dog was excruciatingly slow to respond, waited for five shouts or stopped for an extra wazz on the way back.
Errors are Gold
The type of error is key information as to where the future learning should be focussed: lack of skill, lack of experience, fatigue or simply beyond capabilities at this stage.
The reward is feedback on the quality of the achievement and the work-effort invested. When achievement is less than expected our reward needs to change to encourage awareness of the difference. There must be evidence of competency in these conditions before we include the likelihood of error. An attempt to hide error limits the chance to improve the learning. If an error is carried forward it is likely to magnify with the increase in complexity or difficulty.
Become self-critical of what you pay for
Some dogs are quite capable of deliberate error: checking whether you are awake or just sleeping through the habit. Effort required to perform the behaviour reduces with competency: consider that the reward value should reflect this. The more we practise, the easier it becomes; we should increase the difficulty required for achieving the same reward value. This will be dependent on your assessment of “easier”.
“The purpose of learning is to seek more and better rewards” W. Schultz
There should be no fear of error just opportunities to learn. If we see the error rate begin to increase this may be a wake-up call that you, or the dog, need a break, a good night’s sleep or some extra adoration.
“Rewards are not defined by their physical properties
but by the behavioral changes they induce.” W. Schultz.
Yes, it is complex and dynamic. It may be a change from your previous understanding. But that is the point: the more we learn the better the rewards will be. We may need to rely on external observation and assessment to become aware of our own reward-sleeping and take the nudge to step up, move on an do better. For our dogs the same reward principles apply but we need to be careful of their expectations of rewards, their own evaluation of the effort they invested and whether the reward was worth it. Monitor the changes, make no assumptions.
I love this post! It’s so well explained. I’ve never been concerned when a client’s dog or my own makes mistakes, but I realize I need to explain this more clearly to my clients. This post really made me reflect on that, and I’ll definitely revisit it later. THANK YOU
Thanks for the feedback, error logs are important and an entire subject of their own.