JOY FULL Horses: Understanding Extinction: Part 12

Mastering Micro: Building Unlikely Behaviors with Resurgence
Nothing is either all good or all bad.

We want to use positive reinforcement with our animals because we see it as being both effective and more humane.  But the associations created through positive reinforcement can create addictions to harmful behaviors.  Think about the way advertisers manipulate our behavior to encourage smoking or overeating.

Resurgence and regression can be very negative procedures, but they can also be used to produce what might otherwise be very difficult behaviors to obtain.

If you aren’t sure how you can turn what seems like a negative procedure into a positive teaching strategy, PORTL can once again help to illustrate how this works.

Here’s the set up:

The trainer sets a toy chair on the table for her learner to interact with. The goal is to get the learner to push the chair over the table the way she might push a toy car.

We’ll now observe quietly in the background while the learner begins to interact with the chair.  The trainer could get lucky.  The learner might begin offering the behavior she’s after within the first couple of clicks.  But with this learner there’s no sign of any chair pushing behavior. Why?

History matters.

The learner is going to draw on all of her previous repertoire of things she has done with chairs.  In this case we have a learner who was scolded as a child for pushing her chair over the floor, so she’s not very likely to offer this type of behavior with the toy chair.

A history of punishment has played a role in depressing chair pushing behavior for this learner, but pushing would also have been an unlikely behavior if the trainer had set down a dice. The learner would have tossed the dice or shaken it in her hand because that’s what you do with this kind of object. Pushing a dice over the table like a toy car is not an obvious behavior to try.

Through a series of small approximations, the trainer could try to shaping the behavior she wants.  Her first step would be reinforcing the learner for touching the chair.

The learner in this case is not particularly creative.  She offers simple touches, but nothing else.  Again, the trainer may be dealing with a history of punishment.  Her learner doesn’t have a lot of experience being reinforced for trying things.  In fact, quite the opposite – she may have been punished for stepping “outside the lines”.  She is like so many of our animal learners – hesitant, lacking in confidence, and not showing any outward signs of curiosity.  In her first few attempts she touches the chair, but she doesn’t try any other behaviors.  Getting her to push the chair is going to be hard.

So the trainer takes the chair away and sets out a toy car. Using an object that normally would be pushed makes it very easy to get the desired action.  The learner pushes the car over the table top. Click and treat.

This is repeated several times, and then the trainer takes the car away and sets the chair out.  The learner goes back to touching it.  The chair accidentally falls over – click and treat. The learner latches on to that, expanding her repertoire to two behaviors – touching the chair and knocking it over.

We see this so many times with our animal learners.  One click and suddenly you’ve locked in a behavior you don’t want.  With a creative learner this isn’t a problem.  You can quickly shift the behavior into something you want, but with these “one trick ponies” you have to be so very careful what you click.  In this case the learner persists in knocking the chair over even when she is no longer getting reinforced for the action.

Her trainer makes a quick decision and decides to put everything but pushing the chair like a car on extinction.  Her learner is clearly becoming frustrated.  To avoid having her shut down completely, the trainer takes the chair away and sets the car out again.  The learner immediately starts pushing the car over the table top.  Click and treat.

To help with the generalization the trainer puts a third object out – a small block. The learner pushes the block.  Click and treat.  This is repeated several times, then the trainer takes the block away and sets out the car.  The car is pushed. Click and treat.

The trainer sets the chair out, and the learner pushes the chair.  Job done.

Resurgence and Dog “Yoga”
Using the car in this way is an elegant teaching strategy.  Often when we come up with these clever ways of helping our learner to be successful, we know that it works, but we don’t really have good explanations for why.   Understanding resurgence helps us with the why in this case.  And it helps us to be more deliberate in the use of this kind of teaching strategy.  Here’s another example.

One of Kay Laurence’s students taught her dog to step up with his hind legs onto a chair.  It was elegant training, a beautiful example of setting the learner up for success.  In his talk on extinction, Dr. Jesús Rosales-Ruiz helped us to see that it was also a great example of using resurgence.

Here’s the lesson: First, the dog learned to stand one foot each on four small plastic pods. This alone was impressive training.  The pods were the same ones physiotherapists use to help people improve their balance and proprioception. It took great coordination for the dog to stay balanced on the four pods. But that was only step 1.  Next he learned to keep his front feet on the floor while he maneuvered his hind feet up onto the brick ledge of a fireplace hearth.

Adding in the precision of the pods came next.  Now the dog wasn’t just standing with his front paws on the floor and his hind end up on the ledge.  He was also balancing on all four pods.

This was not done as a cute party trick.  The dog’s owner is a yoga teacher.  Her interest was very much the same as mine – helping her animal learner maintain a healthy spine.  In this orientation she could ask her dog for weight shifts that contribute to a flexible spine.

The last step was setting up a training session next to a chair. The handler withheld the click, putting the dog into an extinction process. With very little experimentation, the dog oriented himself so his hind end was to the chair.  He certainly demonstrated the flexibility of his spine by stepping up onto the chair with his hind legs so he was standing hind end up on the chair and front feet on the floor.

Generalization and Creativity
Jesús commented that if we didn’t know about resurgence we would simply be saying the dog generalized.  That’s not a sufficient explanation.  What we were seeing was a great example of resurgence. PORTL has given us a better understanding of how to encourage this kind of problem solving.  When we want to train for this type of generalization, knowing about the “why” of resurgence helps us to be more deliberate and efficient in our training.

It isn’t positive reinforcement by itself that creates a positive learning experience.  An eagerness for learning comes from being a successful puzzle solver.  That success in turn comes from the kind of efficient, clean training that the clever use of resurgence encourages.

These examples give us a great perspective on creativity.  When we’re training, we aren’t waiting and waiting for our animals to do something we can reinforce.  Instead we can “seed” the behaviors we want them to draw on.  Then we set up the conditions and let them have the pleasure of discovering for themselves new or unlikely combinations.

We have a procedure for setting up the creative process.  You give your learner the repertoire, the components that form more complex behaviors, and then you set a puzzle and let extinction be the catalyst for solving it.

Coming Next: The “Pose”

Remember, if you are new to the JOY Full Horse blog, click on the JOY Full Horses tab at the top of this page to find the full table of contents and links to each of the articles I have published so far.

I hope you will want to share these articles by sending links to this blog to your friends.  But please remember this is copyrighted material.  All rights are reserved. Please do not copy any of the “JOY Full Horses” articles without first getting written permission from Alexandra  Kurland, via theclickercenter.com

Also note: these articles are not intended as an instruction guide for introducing your horse to clicker training.  If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

JOY FULL Horses: Understanding Extinction: Part 9

Eureka Moments: What is Insight?

Using resurgence – Insight
Yesterday I shared several PORTL games developed by Dr. Jesús Rosales-Ruiz.   The games deliberately used extinction.  What was observed was this: when you have been consistently reinforcing behaviors as you establish them in repertoire, and you then remove all reinforcement for them, you get a resurgence of these previously reinforced behaviors.  They reoccur in the order in which they were trained.  

When you instead extinguish the individual behaviors during the teaching phase, you get a different result.  The student will go back to the most recently learned behavior.  If that doesn’t work, he’ll go a little further back, and then a little further back.

In resurgence the behaviors occur in the order in which they were taught, so the oldest behavior in the cluster occurs first.

In regression the order reverses.  The most recently taught behavior reappears first.

So how does this help us?  How can we use this understanding to shape behavior?  To get the ideas rolling Jesús shared several video examples where resurgence was used to train complex, creative behaviors.

The first video came from Robert Epstein’s work. Epstein was B.F. Skinner’s last graduate student.  Together they were exploring the concept of “insight”.  How do we solve puzzles?  Are we truly creating something that has not existed before, or is creativity a product of combining known components to solve a novel puzzle?

Bird Brains
To explore this question Epstein taught a pigeon three component behaviors: pecking a banana, climbing on a box, and pushing the box towards a target.

The pigeon was then put into a chamber with the box and the banana.  The banana was hung up out of reach.  The pigeon couldn’t peck the banana, so an extinction process began. There was a resurgence of previously trained behaviors.  The pigeon was able to push the box under the banana, get up on the box, and peck the banana.

How did the pigeon solve this puzzle so quickly?  What is insight? What really is creativity?  Skinner and Epstein would say the pigeon could solve the problem because it had in its existing repertoire the necessary components.  Pigeons that had no experience pushing the box or jumping up on the box failed to solve the puzzle.

What is Creativity?
Jesús gives us a very process-oriented way thinking about this experiment.  This kind of complex puzzle solving was achieved through resurgence.  Set up the underlying components well, add in a bit of extinction, and “creativity” pops out.

If you leave out one of the components, the individual will struggle to solve the puzzle.  He will experience a much longer extinction process.  Macro extinction emotions will begin to surface, and you have to hope the subject has the persistence to become truly creative.

This is the kind of creativity that is truly stressful.  It’s much better to analyze the end goal – the complex behavior you want to train – break it down into all of it’s component tasks, and then train each of the components separately.  The result will be brilliant looking pigeons that solve in minutes what we might otherwise think would be an impossible puzzle for them.

Persistence
Jesús’ comment was there is “nothing new under the sun”. The behaviors you try are all built out of things you’ve done before.  All the components of what appears to be a novel behavior have been trained in the past. So let’s consider what happens when a group of people are presented with a challenging puzzle.  When they begin experimenting and find that the usual, familiar things aren’t working, some will give up quickly.

Others will persist.  They will experiment with novel combinations of what they already know, but again most will quit if they don’t come up with a solution fairly quickly .

A few will keep trying until they stumble across a novel combination that works.  We call these people inventors and creators because they are persistent enough to find these novel combinations.  The discovery process can be a painful one, but once the new combination has been found, it’s easy for everyone else to copy the results.

I can absolutely relate to this.  Give me a horse puzzle to solve, and I can be very persistent. My life experience has taught me that persistence pays off.  But put me in front of a computer that isn’t cooperating, and I shut down fast. There my experience has produced a different set of expectations. I’ve been in enough situations where errors in a software program have made a problem unsolvable, at least for my level of computer skills.  I don’t have the programing background that makes wrestling with a software issue fun.  Extinction has gone too far and been too uncomfortable.  So in one situation I can be very persistent and creative.  In another I’m the one going through the classic cycle of emotions that macro extinction produces.

I know first hand both how much fun the creative process can be when the expectation of success is there.  And I also know how painful and unpleasant the extinction process is when that expectation is missing.

What I want to create for my learners is a feeling of confidence.  Whether horse or human, I want them to KNOW they can solve whatever training puzzle I throw at them. Build this expectation in early before others have taught them hard lessons about failure, and you get brilliant, enthusiastic, joyful individuals.  They are the optimists of this world.  Whether horse or human, they are fun to be around.  That’s what an understanding of these concepts helps us to create.

Coming Next: Degrees of Freedom

Remember, if you are new to the JOY Full Horse blog, click on the JOY Full Horses tab at the top of this page to find the full table of contents and links to each of the articles I have published so far.

I hope you will want to share these articles by sending links to this blog to your friends.  But please remember this is copyrighted material.  All rights are reserved. Please do not copy any of the “JOY Full Horses” articles without first getting written permission from Alexandra  Kurland, via theclickercenter.com

Also note: these articles are not intended as an instruction guide for introducing your horse to clicker training.  If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

JOY Full Horses: Understanding Extinction: Part 8

Mastering Extinction
Extinction happens all the time.  When you withhold your click, you set up an extinction process.

If you are unclear about your criteria or clumsy in your handling skills, you could be setting up your learner for a macro extinction process with all of the painful emotions that go along with it.

Or you could be using a micro extinction strategy to help shape a more complex behavior.  In this case you are using extinction to your advantage.  Extinction doesn’t have to be something you avoid.  It can be something you actively use to create more complex behavior patterns.

In yesterday’s post I described the PORTL games that Dr. Jesús Rosales-Ruiz  uses to help his students understand principles of behavior.  In his talks he shares some fascinating PORTL experiments to illustrate the difference between resurgence and regression.

Experiment One: Resurgence
The learner was taught a series of behaviors:

Behavior 1: tapping a small block. Once that behavior was confirmed, the block was removed and a toy car was placed on the table.

Behavior 2 was rolling the toy car over the table top.  When the car was brought out for the first time, there was a small extinction burst of tapping the car, but the learner quickly shifted to pushing it.  Pushing a car is an easy guess for what you would do with this kind of object.

When that behavior appeared to be solid, the car was removed and a third object, a key, was placed on the table.  Now the behavior was lifting.  Fingering a key is a normal response to this kind of object so it was easy to get the learner first to touch the key and then to lift it up off the table.  Once the learner was consistently lifting the key, that object was removed and a fourth one was introduced.

Behavior 4 involved the learner putting a wooden ring on her finger.  The learner quickly figured this out and began to consistently offer this behavior.

When each of these behaviors seemed solid – tapping the block, pushing the car, lifting the key, putting a ring on her finger – the trainer reviewed, one at a time, what the learner was to do with each of the objects.

The trainer then placed all four objects out on the table, but not in the order in which they had been taught.  The trainer observed the learner’s behavior.  She did not give any feedback or reinforcement of any kind.  The point was to see in what order the learner would interact with each object.

The result:  The learner went first to object 1/behavior 1, then moved to object 2/behavior 2, then object 3/behavior 3/and finally object 4/behavior 4.

So even though that wasn’t the left to right order in which the objects were set out, that was the order in which the learner interacted with them.

The conclusion: when you have not gone through an extinction process for the behaviors you are using, when you have instead reinforced them, and then you remove reinforcement, you get a resurgence of these previously reinforced behaviors.  They reoccur in the order in which they were trained.  

Now here’s the fun part.  When you instead extinguish the individual behaviors, you get the opposite result.  Now you see regression.  The individual will go back to the most recently learned behavior.  If that doesn’t work, he’ll go a little further back, and then a little further back – thus revealing his training history.

In resurgence the behaviors occur in the order in which they were taught, so the oldest behavior in the cluster occurs first.

In regression the order reverses.  The most recently taught behavior reappears first.

These differences are illustrated in the second experiment.

Experiment Two: Regression
After a series of behaviors have been learned, this experiment again puts the learner through an extinction process.  In the initial set up each time the learner is moved on to a new task, an extinction process is used to eliminate the previous behavior.  Here’s the experiment:

The trainer sets out one item on the table.  The learner begins to manipulate it, trying to find out what is going to be clickable.  The trainer doesn’t click any of this creativity. She waits instead for it to extinguish and then clicks for one simple behavior – touching the object with one finger. That is the “hot” action.

The trainer clicks and reinforces for successful approximations until she has achieved a high degree of consistency in touching the object with one finger.

This was the set up for the experiment.  In the next phase she sets ten different objects out in a circle, including the one they had just been working with.  The learner begins by touching the familiar object.  That gets clicked and reinforced several times, then the trainer stops reinforcing for that object.  She is using extinction to eliminate that behavior.  The learner begins by experimenting, touching various objects, but she only gets clicked for touching the one that was immediately next to the previously hot object in a counter clockwise direction.

The learner switches over to this object and begins touching it consistently.

So now the handler stops reinforcing for this object and only reinforces for the next object on the circle.  The learner again experiments and then discovers that the only object that she gets paid for touching is the third one on the circle.

When this is consistent, the handler again stops reinforcing for touching this object.  The learner is catching on to the overall pattern. Now she moves more quickly to the fourth object and discovers that is the “hot” one to touch.

They continue counter clockwise around the circle until every object has been the “hot” one once and touching it has also been extinguished.

At this point the handler stops reinforcing altogether and simply observes the learner’s behavior.  The result: the learner quickly switches to moving clockwise around the circle, touching the objects in the reverse order in which she learned them.  So she learned them originally counter clockwise: object 1, then object 2, then object 3, then object 4, etc.

Now she was touching them clockwise: object 10 – object 9 – object 8 – object 7, etc.  She isn’t getting clicked for any of these touches, but the pattern is very persistent.

So again: in the first experiment where the behaviors were taught, but not extinguished, the learner went through them in the order in which they had originally been learned.

In the second experiment where behaviors were extinguished, the learner went through them in the reverse order.

You won’t find these distinctions in the scientific literature. These two extinction outcomes, resurgence versus regression, are something Jesús and his students have been revealing by playing PORTL games.

Mind Games
Again Play is the key here.  PORTL may have a serious purpose behind it, but these are games.  All the creativity that comes with play is woven into these experiments.  It may turn out that others playing with similar set ups will have different results.  That’s a good thing.  That simply raises more questions, more puzzles to solve.

Do you have a question about how something works? Great. Design an experiment, test it a few times to work out the kinks in the procedure, and then invite your friends over for a pizza and PORTL party.  In the course of an evening you could have enough data to write a paper!

I do like the new twist Jesús has given to this version of the training game.  As he has pointed out, we’ve been using lab rats to learn about human behavior.  Now we are using humans to model animal behavior. Turnabout is fair play.  Much better to frustrate an undergrad than some poor lab rat!

Coming Next: Eureka Moments!  What is Insight?

Remember, if you are new to the JOY Full Horse blog, click on the JOY Full Horses tab at the top of this page to find the full table of contents and links to each of the articles I have published so far.

I hope you will want to share these articles by sending links to this blog to your friends.  But please remember this is copyrighted material.  All rights are reserved. Please do not copy any of the “JOY Full Horses” articles without first getting written permission from Alexandra  Kurland, via theclickercenter.com

Also note: these articles are not intended as an instruction guide for introducing your horse to clicker training.  If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

JOY FULL Horses: Understanding Extinction Part 7

The Training Game
I’ve mentioned training games several times.  The original clicker training game was a close cousin to the children’s game “Hot and Cold”.  The learner was sent out of ear shot while the rest of the group chose a goal behavior.  When the learner returned, the only instructions she was given were to offer behavior.  If she did something that her designated trainer liked, she would be clicked. She was then to go to her handler for a treat.

I’ve seen situations where the learner got the behavior seamlessly.  One easy click after another led the learner directly to the goal behavior.  I’ve seen other situations where the same behavior tripped people up completely.

When we train our animals, we want the first scenario – seamless, successful training.  That’s what we want for our equine learners.  But in the training game, we often learn the most when we experience clumsy shaping.  It can be frustrating to struggle through a session that lacks a clear training plan, but you do gain a great appreciation for what NOT to do.

Genabacab
Kay Laurence developed a different style of training game.  In this one trainer and learner are seated opposite one another at a table.  Instead of acting out the behavior like a game of charades, the learner manipulates objects which the trainer has set out on the table.

alex-genabacab-with-caption

Kay always has great fun collecting objects for the table game.  She has small plastic fruits and cakes, toy cars, small cones, plastic insects of various varieties.  It’s a colourful mixture that she hands over to her trainers.  When I play the table game at clinics, I raid the host’s kitchen junk drawer.  My toys aren’t as much fun as Kay’s, but they serve the purpose just as well.

Kay calls her game Genabacab.  It has very few instructions and really only one rule: the only person who is allowed to talk is the learner. The trainer and spectators are not to give any verbal hints or to discuss what is going on until afterwards.

The table game lets you work out shaping plans BEFORE you go to your animal.  Do you want to learn how to attach a cue to a behavior and then change that cue to a new cue? You can work out the process playing the table game and spare your animals the frustration of your learning curve.

Kay has described workshops at her training center where someone arrives with a “how do I teach this?” type of question.  Maybe the handler wants to teach match to sample, or she wants to see if her dog can indicate which object is bigger or smaller.  Instead of going straight out to the dog and confusing it with missteps and false starts, everyone in the group will pull out their Genabacab games. Kay says people will often spend half the day happily absorbed in developing the best teaching strategies for their dogs.  The dogs spend the day relaxing while their people work away at the puzzle.  It’s only once the process is well understood, that the dogs are brought in for training.

PORTL
Dr. Jesús Rosales-Ruiz and his students at the University of North Texas have been using Genabacab to understand basic principles of behavior.  He wants to bring the game to the scientific community as a research tool, so he gave his version a new name:  PORTL – Portable Operant Research and Teaching Laboratory.   Kay still has her Genabacab for teaching her canine handlers and Jesús has PORTL for teaching behavior analysis.  On the surface they are similar games, but they serve different functions.

Animal studies are increasingly difficult to do because of ethical concerns and expense.  PORTL offers an alternative for research.  You can have a question about how a particular process works, design an experiment using the PORTL game, and in hour’s time have gathered enough data to write a paper – all without frustrating a single lab rat. Now that’s progress!

His students meet on a regular basis to play PORTL games. When they turned their attention to the extinction process, they made some interesting discoveries.

In one game, the learner was shaped to place one hand over the other – right hand over left, and then to reverse it – left hand over right.  The behavior was put on a fixed ratio of 5, meaning the learner was clicked and reinforced on every fifth swap of hands.

The second task was tapping a block.  Again the learner was put on a fixed ratio of 5. (The learner was to tap the block five times for each click and treat.)

The trainer then increased the ratio for the tapping to 30. The learner began to tap the block, but now there was no click and treat after 5 taps.  The learner kept going to about 13 taps.  At that point she began to experiment.  She reverted back to swapping hands.  Then she tried a few more taps, before going back to hand swaps.  She tapped the block a few more times.  The trainer was still keeping track so each of these taps was counting towards the count of 30 she was looking for.

In the twenties the learner began to be creative.  She tried different ways to move hand over hand.  She’d go back and forth between experimenting with hand swaps and tapping the block.  Finally she reached a count of 30 at which point her handler clicked and reinforced her.  All the extra gunk was also chained in.  Now as the handler kept reinforcing the tapping of the block, the frequency of the hand swapping also skyrocketed.  That behavior was no longer being intentionally reinforced, but it increased right along with the tapping.

Now you may be thinking:  “Well that’s just poor training.  No one is going to jump from a fixed ratio of 5 to one of 30.” My response would be to say that this can happen inadvertently.

Suppose a handler has had a behavior on a high rate of reinforcement. The horse is responding on a consistent basis, but then he’s distracted. He’s no longer offering the same consistent response.  Instead the handler is seeing a string of unwanted behaviors.  Sometimes the horse almost meets criterion, but not enough to click. And then he comes through with the right answer.  The handler captures that moment with a click and a treat.  The question is: what is the long term result of that click? Has the handler just identified a single clickable moment or has she chained in a long string of “junk” behavior?

The horse’s future responses will answer that particular question, but Jesús’ response in general is: if you want clean behavior, you need to train in clean loops.  Kay and I would add that you need to microshape.  You need to learn to set up your training so the behavior you want is the behavior you get.

Here’s a link to a great youtube video of  a PORTL game presented by Mary Hunter.   Many of you will know Mary from her StaleCheerios.com blogs. Mary is president of The Art and Science of Animal Training, the organization that puts on the annual conference of that same name in Dallas TX.  She and Jesús will be presenting a program on PORTL at this year’s clicker Expos.

Coming Next: Mastering Extinction

Remember, if you are new to the JOY Full Horse blog, click on the JOY Full Horses tab at the top of this page to find the full table of contents and links to each of the articles I have published so far.

I hope you will want to share these articles by sending links to this blog to your friends.  But please remember this is copyrighted material.  All rights are reserved. Please do not copy any of the “JOY Full Horses” articles without first getting written permission from Alexandra  Kurland, via theclickercenter.com

Also note: these articles are not intended as an instruction guide for introducing your horse to clicker training.  If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

5GoToSea: Pt 11: Mastering Extinction

Resurgence and Regression: Understanding Extinction So You Can Master It

From a presentation given by Dr. Jesús Rosales-Ruiz during the 2014 Five Go To Sea Conference cruise.

Part 1: The Elevator Question
Part 2: The Translation to Horses: Is Personality Expressed or Suppressed?
Part 3: Unraveling the Regression Mess
Part 4: Extinction and Shaping
Part 5: Extinction Reveals The Past
Part 6: Accidental Extinction
Part 7: Emotions
Part 8: Training With High Rates Of Reinforcement
Part 9: Cues and Extinction
Part 10: PORTL
Part 11: Mastering Extinction

If you have not yet read the previous articles, I suggest you begin with Part 1.

Part 11: Mastering Extinction

Using PORTL to Master Extinction
Extinction happens all the time.  When you withhold your click, you set up an extinction process.

If you withhold the click because you are unclear about your criteria or you’re clumsy in your handling skills, you could be setting up your learner for a macro extinction with all of the painful emotions that go along with it.

Or you could be withholding the click with very deliberate intent.  In this case you are using a micro extinction process to help shape a more complex behavior.  You are using extinction to your advantage.  The conclusion: extinction doesn’t have to be something you avoid.  It can be something you actively use to create more complex behavior patterns.

Today’s post shows how the shaping game, PORTL can help us understand how this works.

Jesús watching participants at a conference learning to play PORTL.

Jesús coaching participants at a conference as they learn how to play PORTL.

Shaping with Resurgence
Jesús used videos of PORTL experiments to illustrate what he meant.  He reminded us that there is a difference between resurgence and regression.  The first video example showed an elegant use of resurgence.

The learner was taught Behavior 1: tapping a small block.  Once that behavior was confirmed the block was removed and a toy car was placed on the table.

 

Behavior 2 was rolling the toy car over the table top.  When the car was brought out for the first time, there was a small extinction burst of tapping the car, but the learner quickly shifted to pushing it.

When that behavior appeared to be solid, the car was removed and a third object was placed on the table.  Now the behavior was lifting.

Behavior 4 was a different action.  The learner put a wooden ring on her finger.

When each of these behaviors seemed solid, the trainer reviewed one at a time what the learner was to do with each of the objects.

The trainer then placed all four objects out on the table but not in the order in which they had been taught.  There was now no reinforcement given.  The trainer was simply observing the learner’s behavior – not giving any feedback or reinforcement of any kind.  The point was to see in what order the learner would interact with each object.

The result:  The learner went first to object 1/behavior 1, then moved to object 2/behavior 2, then object 3/behavior 3/and finally object 4/behavior 4.

So even though that wasn’t the left to right order in which the objects were set out, that was the order in which the learner interacted with them.

The conclusion: when you have not gone through an extinction process for the behaviors you are using, when you have instead reinforced them and then you remove reinforcement, the behaviors occur in the order in which they were trained.  This is resurgence as opposed to regression.

Shaping with Regression
Now here’s the fun part.  When you first extinguish the individual behaviors, you get the opposite result.  Now you see regression.  People will go back to the most recently learned behavior.  If that doesn’t work, they’ll go a little further back, and then a little further back – thus revealing their training history.

Jesús showed a second video, this one was exploring what happens in a shaping session where you reinforce an approximation, and then go through an extinction process so you can switch to a new behavior.  Here’s the set up:

The trainer set out one item on the table.  The learner began to manipulate it, trying to find out what was going to be clickable.  The trainer didn’t click any of this creativity.  She waited for it to extinguish and then clicked for one simple behavior – touching the object with one finger.  That was the “hot” action.

The trainer clicked and reinforced for successful approximations, then she took a break to record her data.  She continued to train in ten click units until she had achieved a high degree of consistency in touching the object with one finger.

This was the set up for the experiment.  In the next phase she set out in a circle nine or ten different objects, including the one they had been working with.  The learner began by touching the familiar object.  That got clicked and reinforced several times, then the trainer stopped reinforcing for that object.  She was using extinction to eliminate that behavior.  The learner began experimenting, touching various objects, but she only got clicked for touching the one that was immediately next to the previously hot object in the counter-clockwise direction.

The learner switched over to this object and began touching it consistently.

So now the handler stopped reinforcing for this object and only reinforced for the next object on the circle.  The learner again experimented and then discovered that the only object that she got paid for touching was the third one on the circle.

When this was consistent, the handler again stopped reinforcing for touching this object.  The learner was catching on to the overall pattern.  Now she moved more quickly to the fourth object and discovered that was the “hot” one to touch.

They continued counter-clockwise around the circle until every object had been the “hot” one once and touching it had also been extinguished.

At this point the handler stopped reinforcing for anything and observed the learner’s behavior. The result: the learner quickly switched to moving clockwise around the circle, touching  the objects in the reverse order in which she had learned them.  So she learned them originally counter-clockwise: object 1, then object 2, then object 3, then object 4, etc.

Now she was touching them clockwise: object 10 – object 9 – object 8 – object 7, etc.  She wasn’t getting clicked for any of these touches, but the pattern was very persistent.

What PORTL Reveals
So again: in the first video where the behaviors were taught, but not extinguished, the learner went through them in the order in which they had originally been learned.

In the second video where behaviors were extinguished, the learner went through them in the reverse order.

You won’t find these distinctions in the scientific literature.  This difference in behavior relating to resurgence and regression is something Jesús and his students have been revealing by playing PORTL.

Coming Soon: Part 12: Creativity Explored

Please note: If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

5GoToSea: Pt 10: PORTL

Resurgence and Regression: Understanding Extinction So You Can Master It

From a presentation given by Dr. Jesús Rosales-Ruiz during the 2014 Five Go To Sea Conference cruise.

This is Part 10 of a 15 part series.
Part 1: The Elevator Question
Part 2: The Translation to Horses: Is Personality Expressed or Suppressed?
Part 3: Unraveling the Regression Mess
Part 4: Extinction and Shaping
Part 5: Extinction Reveals The Past
Part 6: Accidental Extinction
Part 7: Emotions
Part 8: Training With High Rates Of Reinforcement
Part 9: Cues and Extinction
Part 10: PORTL

If you have not yet read the previous articles, I suggest you begin with Part 1.
Part 10: PORTL

PORTL’s Origins

Jesús and the son of one of the conference attendees playing PORTL during the Five Go To Sea cruise.

Jesús and the son of one of the conference attendees playing PORTL during the Five Go To Sea cruise.

PORTL evolved out of Genabacab, a table game Kay Laurence developed for teaching shaping.  Genabacab has very few instructions and really only one rule: the only person who is allowed to talk is the learner.  The trainer and spectators are not to give any verbal hints or to discuss what is going on until afterwards.

The table game lets you work out shaping plans BEFORE you go to your animal.  Do you want to learn how to attach a cue to a behavior and then change that cue to a new cue?  You can work out the process playing the table game and spare your animals the frustration of your learning curve.

Kay has described workshops at her training center where someone arrives with a “how do I teach this?” type of question. Maybe the handler wants to teach match to sample, or she wants to see if her dog can indicate which object is the biggest one in a set. Instead of going straight out to the dog and confusing it with missteps and false starts, everyone in the group will pull out their Genabacab games.  They will spend hours happily absorbed in developing the best teaching strategies for their dogs.  Their dogs, meanwhile, are spending the day relaxing while their people work away at the puzzle. It’s only once the process is well understood, that the dogs are brought in for training.

Jesús has been using Genabacab to help his students understand the concepts of learning theory.  He wants to bring the game to the scientific community as a research tool, so – with Kay’s blessing – he has renamed it.  It is now PORTL – Portable Operant Research and Teaching Laboratory.  Animal studies are increasingly difficult to do. They are expensive, and there is always the question of ethics.  How fair is it to run studies on lab rats?  PORTL is a much better solution.  You can have a question about how a particular process works, design an experiment using the PORTL game, and in the course of an evening have gathered enough data playing the game with a group of undergrads to write a paper – all without frustrating a single lab rat.  Now that’s progress!

PORTL Games
Jesús’ students meet on a regular basis to play PORTL games.  In his talk he showed some videos that illustrated beautifully how he used it to ask questions about regression and resurgence.  In one video two tasks were taught.  First, the learner was shaped to place one hand over the other – right hand over left, and then to reverse it – left hand over right.  The behavior was put on a fixed ratio of 5.  That means the learner was clicked and reinforced on every fifth swap of hands.

The second task was tapping a block.  Again, the learner was put on a fixed ratio of 5. (The learner was to tap the block five times for each click and treat.)

The trainer then increased the ratio for the tapping to 30.  The learner began  to tap the block, but now there was no click and treat after 5 taps.  The learner kept going to about 13 taps.  At that point she began to experiment.  She reverted back to swapping hands.  Then she tried a few more taps, before going back to hand swaps.  She tapped the block a few more times.  The trainer was still keeping track so each of these taps was counting towards the count of 30 she was looking for.

In the twenties the learner began to be creative.  She tried different ways to move hand over hand.  She’d go back and forth between experimenting with hand swaps and tapping the block.  Finally she reached a count of 30 at which point her handler clicked and reinforced her.  Jesús’ point is now all the extra gunk was also chained in. If the handler were to keep reinforcing the tapping of the block, she would also see the frequency of the hand swapping skyrocket.  That’s not the desired, goal behavior, but it would increase right along with the tapping.

Now you are probably thinking:  “Well that’s just poor training.  No one is going to jump from a fixed ratio of 5 to one of 30.”  My response would be to say that this can happen inadvertently. Suppose a handler has had a behavior on a high rate of reinforcement.  He’s asking the horse to carry himself in a correct bend.  He’s cueing it through gentle requests down the rein.  The horse is responding on a consistent basis, but then he’s distracted or the footing changes so he loses his balance. Whatever the reason, the handler isn’t getting the same consistent response. Instead he’s getting a string of unwanted behavior.  Sometimes the horse almost meets criterion, but not enough to click.  And then he comes through with the right answer.  The handler captures that moment with a click and a treat.  The question is: what is the long term result of that click?  Has the handler just identified a single clickable moment or has he chained in a long string of unwanted behavior?

The horse’s future responses will answer that particular question, but Jesús response in general is: if you want clean behavior, you need to learn to microshape.

Jesús made the further comment that this type of inadvertent chaining happens all too often when people are working with autistic children.  The sad thing here is the previous behavior that is being reinforced is not something harmless like hand swapping, but it’s often self-injurious behavior like head banging.

To sort out the tangle you need to analyze the whole behavior chain rather than focus in on individual behaviors.  These adjunctive behaviors can create a lot of stress. Again Jesús emphasized that’s why it is so important to understand extinction.  You need to understand it so you can master it.

Coming Soon: Part 11: Mastering Extinction

Please note: If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com

5GoToSea: Pt. 2: The Translation to Horses

Resurgence and Regression: Understanding Extinction So You Can Master It

aftercruise1From a presentation given by Dr. Jesús Rosales-Ruiz during the 2014 Five Go To Sea Conference cruise. 

This is Part 2 of a 15 Part series.

Part 1: The Elevator Question
Part 2: The Translation to Horses: Is Personality Expressed or Suppressed?

Part 2: The Translation to Horses: Is Personality Expressed or Suppressed?

Personality Expressed or Suppressed
In the opening of his presentation on regression and resurgence Dr. Jesús Rosales-Ruiz defined regression and gave some examples in terms of human behavior.  I ended yesterday’s post with this statement:

Extinction reveals our history.roman ruins

How does this translate to horse training?  At the very beginning of clicker training the extinction process may reveal your horse’s training history. It helps us to understand the “childhood” our horses have had.  Did your horse have a fair introduction to people, or are there issues you need to know about?

In most cases when you introduce a horse to the clicker, it’s smooth sailing.  The horse quickly figures out the game. You may have to go through a little bit of explaining around the food, but for most horses this moves along without any major hiccups.  You hold a target up, he investigates it, click, you give him a treat.  Easy. Unless he’s one of the horses who has been punished for showing any self-expression.

If your horse has learned that being “well behaved” means he doesn’t offer any behavior you haven’t asked for, he’ll be good at following orders, but not taking the initiative. In fact your “well-behaved” horse may have learned that offering behavior is dangerous.  The best way to avoid punishment is to wait to be told what to do. This is why I put well-behaved in quotes.  Is he well behaved in the way a clicker-trained horse can be?  Or is he simply not offering much in the way of behavior?  There’s a huge difference.  In the first, the personality is expressed.  In the later, it is suppressed.

When you hold out the target, a suppressed horse may be stuck for answers.  He doesn’t know what you want. The “right answers” that normally work don’t seem to apply in this new situation.  This horse is being presented with a puzzle that can make him feel very uneasy.  In the past guessing wrong has meant being punished.

Extinction Reveals the Past
At first this horse may try offering the one or two things that might possibly fit this situation. When those don’t work, he’ll become aggressive. He’s going to protect himself from the punishment he’s knows is coming when he doesn’t respond right away.  Your “well behaved” horse is suddenly charging you with teeth bared.  It’s easy to blame clicker training or the treats for this sudden turnaround in behavior, but I’ve always seen it very differently.  I’ve always said that what is happening is the training history of the horse is being revealed.  Jesús’ talk confirmed this.

Often what we refer to as “well behaved” horses are really horses whose behavior and personality have been shut down through the use of corrections.  They have learned to wait to be told what to do.  Offering behavior, expressing their personality has been punished.  Give them a command they know, and they will respond promptly.  They can seem like the perfect horse.  Safe, easy to direct.  But put them into a situation where they don’t know the answer – in dodo birdfact they really don’t even understand the question – and you will begin to see the extinction process unfold.  Extinction follows a predictable pattern.  These horses will take you back through the stair steps of how they have been treated, and often the story they tell is not a pretty one.

When a horse is not sure of the “safe” answer, he’ll begin to regress back through his training history.  You will see the behavior that has been “swept under the carpet” by suppressing it with corrections.

How do you avoid this regression back into unsafe behavior?  The early steps of clicker training are very structured.  I make use of protective contact so the horse is free to interact – or not.  This lets me see what kind of a learner I have so I can tailor those early steps to the individual. I design my lessons around very small steps so I can keep the training loop clean.  That doesn’t just mean that the horse performs the intended behavior.  Everything matters.  How he takes the treat matters.  How long he hesitates before beginning a new cycle matters.  How quickly he performs the desired behavior matters.  These all tell me something about the emotions he’s experiencing and those definitely matter.  My goal in this first foray into clicker training is to avoid the kind of uncertainty that leads to frustration and a regression back through older learning patterns.

Details matter – especially in shaping.  Jesús showed a couple of video examples of shaping where the loop was not kept clean.  In one a dog was going to be reinforced for coming back to the handler away from distractions.  While the instructor was explaining the lesson, the dog’s handler was listening to her, not paying attention to her dog.

The dog started surfing through all the behaviors that had been reinforced in the past. What should he do to get his person focused back on him?  He started with head bobs, moved on to sitting, then a play bow into lying down and finally he started jumping up on his person.  These were all behaviors that had previously been reinforced, sometimes unintentionally.

The instructor finished describing what she wanted the handler to do, and the formal “session” began.  The instructor deliberately distracted the dog while the handler tried to call him away.  The dog returned fairly promptly to his handler, but the behavior included a sit into a play bow followed by the dog lying down, then jumping up on the handler.  So yes, the dog did indeed return to the handler, but the recall now included these other unwanted behaviors.

This is why I stress so much how important it is to pay attention to details.  When you are first starting, it can be hard to keep track of everything, but details matter.  Yes, you can get your horse touching targets.  Yes, you can have a lot of fun. Yes, clicker training can be very easy.  But if you aren’t being attentive to details, you can miss a lot of important signs that your horse may not be fully understanding this new game.  If your horse isn’t sure what is wanted, you could see a regression through his past training history.  He’ll be telling you what he thought of how he’s been treated, and often the tale is not a pretty one!

When you are brand new to clicker training, and especially if you are also new to horses, this can be a hard dynamic to understand.  What you hear about clicker training is how much fun it is, how much horses enjoy it.  So you give it a try. But instead of smooth sailing, your horse falls apart.  Instead of having a wonderful time, you’re dodging teeth.  You’ve been promised a dream horse, and all you have is a nightmare.  Of course, you blame clicker training and all of the treats you’re feeding for the horrible behavior you’re seeing.   But what can you do?  You don’t want to go back to punishment-based solutions.  You keep hearing from others that you need to trust the process, so that’s what you do.  You continue on determined to solve the riddle of your horse’s regression into nightmare behavior.

Coming tomorrow: What it means to trust the process: Unraveling the regression mess.

Please note: If you are new to clicker training and you are looking for how-to instructions, you will find what you need at my web sites:

theclickercenter.com                    theclickercentercourse.com