Sunday, July 13, 2014

When Failure Is An Option: Why We Have To Let Animals Mess Up Once in a While


Failure is sometimes an option, Gene.  Well, at least in animal training.  Maybe not so much in space.

How does that word make you feel?  For me, it used to make me feel really yucky.  I totally dreaded doing poorly in school because I thought my parents would get mad.  Then as I got older, I thought it meant I'd never get into college, and then I thought I'd never get a job with bad grades.  When I got a job, I thought failure meant I'd never make it as an animal trainer, nor would I climb the career ladder.

At some point though, I realized that I learned a lot from my failures and mistakes.  I slowly learned that failure doesn't always have to be a bad thing.  Maybe it feels rotten at the time, but it can provide a lot of opportunity.  It is sometimes even a badge of honor, because it means you may have gone out on a limb and took a risk.

There are probably a gadrillion (give or take) resources on how mistakes and failure to certain extents make us stronger and better.  And by us, I mean us humans.

But what about the animals we train?  What about when they mess up?

Penguins both mess up AND make a mess.  And boy, do they make big messes.

For those of you who don't know, this blog was named The Middle Flipper because a) I was inspired by the animals who tell me NO WAY MAN and b) an awesome former supervisor of mine used to always joke that a dolphin was giving us the middle flipper whenever they'd refuse to do something.  Side note: if you didn't know why this blog was named the way it was, I hope you have a hilarious story to tell me about what you thought Middle Flipper meant.

Anyways, so I dedicate this entire blog to the animals who say no thanks, and to the concept that the animals have a choice in their training.  But what about when they try...and fail?  What happens then?  Not just simply, "Nah, I don't want to do that" but more like, "Whoa, I can't do that" or "Uh, wait I thought I was right?"

Pick up any book on positive reinforcement training, ESPECIALLY marine mammal training, and you'll see a common phrase like, "always end on a positive" and "set up the animals for success."  Those are amazing concepts and are used a lot, because of course you want the animals to find the training sessions reinforcing.  It's also, I think, a pendulum swing away from the old-school methods of training involving a lot of correction and positive punishment* where animals were routinely set up to fail and were essentially coerced into performance.

Heinz Getwellvet may have used old school animal training methods.

I am a huge advocate of positive reinforcement training, d'uh.  I'm also a big fan of the "set your animal up for success" methodology.  But is it possible to go too far to the "always end on a positive" spectrum?  The answer is yes.

Animals, like us, can learn from failure.  In fact, learning how to mess up is a critical skill.  It's a skill that requires careful and empathetic training, knowing that in the long run, the animal will benefit if they are taught how to think outside the box and deal with occasional mistakes.   You as the trainer must teach the animals in your care how to fail.  And you also have to know when it's okay to not end on a high note, for the sake of the animal's learning curve.  So let's explore these ideas.

First, how do you teach an animal to accept making a mistake?  In many facilities, the LRS is the answer.  For those of you who don't know, the LRS stands for "least reinforcing stimulus" or "least reinforcing scenario".   The LRS is a 3-5 second neutral response following an undesired behavior.  That means if I ask Augustina the seal to wave and she sticks her tongue out at me**, I simply stop giving the wave SD and count to three.  

Here's a cute picture of a seal, because when I typed in "LRS" in google, a bunch of surgical photos came up.  Don't believe me? Try it.  You'll thank me for my photo choice.

Depending on your understanding of operant conditioning terminology (which may or may not be as technical as the vernacular of Trekkies), you may have your own opinion about what an LRS is,  but in practice it is a simple way to tell an animal, "Ehhhh that wasn't quite right, but if you react calmly to this stimulus which basically tells you you were incorrect, we will move on to something reinforcing/fun."

What are they even saying?

One of the key elements of the LRS is the animal's response while it is being administered.  If while I'm LRSing an otter and she decides to take this time to wander off, find an insect in her habitat and pull its legs off one by one, that isn't the correct LRS response.  If a sea lion becomes frustrated with the lack of reinforcement due to the LRS happening and charges me, that isn't the correct LRS response.  Why? Because it doesn't show that the animal is capable of accepting being wrong.  It shows the opposite.

If however, the animal sits calmly at station (whatever criteria you've established) for the LRS, that is the ideal LRS response.  Why?  Because the animal understands that it's worth his/her while to stay in the session, fully attentive.  They have learned, "Eh, it's okay I messed that up.  My trainer's going to help me out."

However, you HAVE to teach the LRS response.  You don't just automatically get it.  Even if you have an animal who is really laid back and doesn't seem to mind not getting everything correct, at some point, if you don't reinforce/maintain the LRS properly, you will end up with an unmotivated or potentially frustrated animal on your hands.

Calvin! How are you supposed to teach Hobbes an LRS response if you can't keep your &@%# together?!

So okay, you might think the LRS spiel is a no brainer.  LRS's are standard in most marine mammal training programs, because it clearly marks undesired behavior without the animal getting all down in the dumps.   But not every place uses them, and they have their reasons.  I've worked with a couple of groups of animals who did not know an LRS.  I've talked to people who don't believe that they are the most effective way at teaching an animal, and that is totally okay because their animals have learned another method (still positive reinforcement) of dealing with mistakes.

But what about when the animal just isn't succeeding?  When your session has taken a turn down Oh God Nothing Is Going Right Lane?

It's time for a time out, or to end the session for a while.  

A time out is usually 30 seconds to five minutes long.  Some facilities use an Extended Time Out or ETO, which is basically when you're like, "okay, let's just come back at the next scheduled session."  

What's the benefit of a time out (including ETOs)? There are a lot, but for the sake of this blog topic, the obvious one is that you communicate that something or series of things were so undesirable that it merits ending the session.  Some of those things may involve animals who displace each other, an animal who just doesn't want to do anything he or she is being asked, or an animal who is being well, naughty.  Like if they keep stealing toys from you or something like that.  Or perhaps they repeatedly refuse to do a behavior that's really solid or is making a lot of progress.

Here's a recent example.  One of our older sea lions learned a voluntary blood last year.  She has been really amazing at it.  The entire behavior is 100% voluntary (as it ought to be), and she at any time can just get up and walk away.  Does that mean her session is over? Sure, maybe for that time.  But she knows because of her behavioral history that she will get many other sessions throughout the day in which time she'll receive all of her food, unless she chooses to not eat it all.

No thank you!

The trainers who have worked on this blood behavior have made it very reinforcing.  They've been spoiled in a way, because this sea lion has been super solid at the behavior, even despite some health issues related to her very advanced age.  The trainers and the sea lion have been on the same wave-length for a while and it's proven to be very beneficial medically for the old gal.

But, as things tend to happen, nothing is perfect. 

A few days ago, we tried for a blood on this animal.  And she chose to say no.  She was very polite; she slowly pulled her rear flipper away from the needle stick, turned her head around towards the trainers and did a little, soft vocal.  

After the LRS, her trainers set her up again.  Again, the sea lion said no thanks.  

At this point, we are at a fork in the road aren't we?  On the one hand, we are thinking, "Oh man, I have to end this session on a positive!!! I can't let this end on a negative, not a behavior that's been so solid and is SO critical!"  But on the other hand we think, "Wait, this sea lion is normally so good at this.  What are we communicating to her if we take a severely regressed approximation?"

So which path do we choose?

Choose your own adventure books are...amazing.

My suggestion is to look at it another way.  Depending on what decision you make, which one will the sea lion learn the MOST from?  We as trainers always say the training is primarily for the animals' benefit, not for our own.  That includes deciding as a trainer (TEACHER!) what is best for the animal's learning, not what's best for compliance, or our ego, or our need for the animal's success.

In this particular case, the sea lion is choosing to say no.  I wouldn't even say it was a Middle Flipper Event, because there was no dramatic refusal.  It was a series of simple, subtle gestures.  But still, she said nope, not today.  Positive reinforcement training is based in empowering the animal to make a choice; to not feel as though they HAVE to do something.  If they want the particular reinforcements we offer in a session, then they'll emit the behavior.  But if what we have to offer isn't enough or what they want in that session, they may choose to decline.

By continuing on with this sea lion just to get a positive, we would (in this case) be appealing to our own egos as trainers.  Not to the animal's best interest.  Perhaps I'd have a different answer if we continued to see this behavior break down in future sessions, but in one session, we don't have enough information to see a pattern.  All we know is that for some reason, she doesn't want to do it NOW.  If we back up and take approximations that are WAY regressed from the solid, maintained blood behavior, what does the sea lion learn?  One of two (and potentially both) things:

1) If she refuses the behavior that she's been great at for almost an entire year, she will still get reinforcement for doing significantly less.  

2) Her refusal is not respected by the trainers.  She is not "allowed" to say no.

If she learns #1, well now we have a consistency problem on our hands.  Do we want her to understand that she has to lay out for the entire blood behavior until she is bridged?  Yes.  Has she been doing that behavior to criteria for a long time? Yes.  So why would we then change the rules just because of one failed session?  Are we positive that this sea lion understands why we are suddenly taking much earlier stages of the behavior after she's just refused to do the full thing?  Absolutely not (unless someone on my staff can speak to sea lions, which would be awesome because I have a lot of questions such as why they bark at their butts when they fart really loudly).

...when you're fast asleep?
More fart shirts here:

If she learns #2, then you've got a big motivation problem.  Letting the animals say "no" in the moment is a tool to significantly reduce or eliminate animal frustration.  Again, please understand I'm not suggesting that you do ANYTHING 100% of the time, but if an animal suddenly stops doing something they are consistent at, you let them say no and end your session.  It's not a bad idea to just give them their space, even if in your head you say it's because you're giving them a time out.

Now what would this sea lion learn if we took a time out?  Essentially the opposite of the aforementioned list.  

1) If she refuses the behavior that she's been great at for almost an entire year, the session is over.

2) Her refusal is respected by trainers.  She is allowed to say no, which means the trainers will go back to the drawing board to troubleshoot the behavior and set her up for future success.

Future success.  That's long-term success, not just the immediate WIN you get by saying "oh, I got a positive on a train wreck of a session".  

So it was understandable when the trainers decided to respect the sea lion's refusal and end the session, that they were bummed about it.  They talked to me about it later, saying, "Man, she is normally so good at that!! It kills me we had to take a time out."

But does that really affect the sea lion the same way?  Probably not.  It bothers us more as trainers/teachers because we want our animals to succeed, and let's face it, we also don't want the reason they messed up to be us (newsflash: sometimes it IS our fault, but that's okay!!).  Of course, it doesn't help when two of your bosses are watching you.  But it's important to realize that your ego cannot be a reason you make a behavioral decision.  The reason must be because it is what teaches the best lesson to the animal in your care.

Luna agrees it's all about her.

Cognition training is a great example of how failures can teach as much as successes.  When I was training a dolphin to do a match-to-sample behavior, where I basically showed her a toy and she had to go find its match in a line up of different toys, I learned this lesson well.  Even though the dolphin obviously did not get every single trial correct, she learned each time she chose an object, right or wrong.  All of her responses and the consequences afterwards (hearing a whistle if she was right, not hearing one if she wasn't) gave her information to make future decisions.

"Ending on a positive" or "set the animal up for success" is a rule of thumb.  We should have WAY more successes than failures.  There's no doubt about that.  We can look to our own preference in learning.  Would you rather be called out on 90% of what you do well, and 10% of what you need to improve on?  Or would you rather be told about 90% of your mistakes, and only 10% of what you do right?  No. Brainer.

But it is so so so so important to remember that we LEARN from mistakes, and so do our animals.  Some mistakes are so big, or come so out of left field that it's BETTER to let the animal learn that lesson than it is to completely confuse them or remove their choice to refuse something just because we think we have to End On A High Note.  We can allow a session to be a catastrophe behaviorally (obviously, not talking about safety scenarios here).

Not all catastrophes are bad.  Some are just adorable.

Hey, I've done it before.  We all have.  Especially when you're really making progress in a certain behavioral scenario, with a new behavior, or with a particular animal.  It SUCKS when a great session (or at least one with great potential) goes down the toilet.   And I am just as guilty as anyone else in trying to get some semblance of the behavior "back on a positive".  Like, if I lay out a dolphin for a blood and they refuse it,  I've definitely fallen into the Trainer Pit Of Despair and called a session "positive" because hey, at least the dolphin let me kind of sort of touch the peripheral molecules of their flukes for 0.000000000001 seconds, instead of what they normally let me do which is GET A BLOOD.

But maybe next time we're in a situation where we think the session is tanking, we think to ourselves, "Is it possible to salvage this session in a way that is meaningful and consistent to the behavioral history of this animal?"  The answer may very well be yes.  In fact, I hope many times it is.  But you also have to be open to the idea that it isn't.  That the best thing is to take a time out or ETO.  

The last idea I'll end on is a phrase I heard a former boss of mine use.  I really like it, because it's simple and very, very true.  

"LRS's do not fix problems, positive reinforcement does."

Also, donuts.

Totally, totally agree.  That concept can still be applied to everything I've just said.  Our sea lion may be better at the blood behavior the next time we do an approximation thanks to us just calling it quits on that one particular session.  But that's not where the troubleshooting ends.  We have to make sure that the elements that appeared to cause the refusal have a high reinforcement history when the sea lion chooses to emit them correctly.  That is what will fix the problem.  But before that can happen in this case, we must first allow the animal to screw up without us freaking out that we ended on a "negative".

I hope I'm being clear on how letting an animal fail once in a while can help him or her learn.  It is a delicate balance and requires experience, confidence, patience, and of course the willingness of ourselves as trainers to make the wrong decision and learn from it ourselves.  Positive reinforcement is a powerful, amazing way to communicate to and train animals to do amazing things.  But we all have to struggle once in a while in order to grow. Even us Homo sapiens.  :)

* If you think this is an oxymoron, just know "positive" doesn't mean "yayayayay", but "adding to".  So basically, adding an aversive stimulus as a consequence to behavior you want to see decrease in frequency.  Example: I don't do the dishes, so someone puts cockroaches on me.  The cockroaches are the positive punisher.  And also they are disgusting. 

** This happens all the time

No comments:

Post a Comment