Friday, January 8, 2010

My Dog is Smarter Than Me, or: Breaking the Reactive Behavior Chain

My dog is smarter than me.

It is often said that whenever you are with your dog, one of you is training the other. Every moment is a subtle yet complex set of interactions in which behaviors are being established, maintained or changed. Sometimes we train the dog, but sometimes the dog trains us. Maisy, in fact, is an excellent trainer, so skilled that I often don't realize that she is shaping my behavior in her favor. She not only has a very clear idea of the behaviors she wants from me, but she also is very good at eliciting the behaviors.

I want you to throw my ball. I want you to give me a treat. What do I have to do to get you to do that? Does this work? No? How about this? No? How about this? Ah-ha! That worked! I’ll do it again! And again! And again!

Of course, this is what good dog training is all about- the dog works for a reward. It’s a win/win situation, really. The handler gets what they want- a dog sitting or coming or whatever- and the dog gets what they want- a treat or a ball. The trick is that the handler has to time the reward to coincide with the behavior she wants; if the timing is off, the dog learns to do something else entirely.

And, I obviously made some mistakes in what I reinforced, because Maisy has figured out that if she displays a reactive behavior, there is a very good chance that I’ll give her a treat.

Here’s how she learned that: in the past, when Maisy would react to something scary by lunging and growling or barking, I would call her name to interrupt the behavior and direct her attention away from the trigger and back to me. I timed my click and treat so that I thought I was reinforcing her becoming silent and looking at me. And I guess I was, because she learned that if she lunged and barked or growled, and then returned to me quietly, she’d get a treat.

Like any dog who has been reinforced for doing something, she started offering that behavior, even if it was a behavior that I didn’t want. She didn't know that. After all, a dog can't know what we intend to train. A dog simply does what works. And acting reactive worked for Maisy.

It took me awhile to realize what was going on. I first noticed that she didn’t seem as scared of things, and that her focus and attention was improving. Then, my amazing trainers, Jane and Robin pointed out that Maisy seemed different in class. She was still displaying reactive behaviors, but her body language didn’t really match up with her actions; she looked loose and relaxed while lunging, and self-interrupt her behavior to turn back to me with this expectant look, as if to say, “Did you see that mom? I did good, huh? Do I get a treat?”

After several weeks of this, I decided that something needed to change. While it was clear that I’d been quite successful in counter conditioning her fear response, I also began to understand that in the process, I had inadvertently created a behavior chain in which she could earn treats by “acting reactive.” After some discussion with Jane and Robin, I decided to take away the rewards so that the behavior didn’t pay off any more (something that is technically called extinction).

So, two weeks ago, I started working with Maisy differently. I decided that I would ignore all future instances of reactive behavior. If she lunges and barks or growls at something, I don’t call her name, I don’t try to interrupt the behavior, I don’t ask her to do something else instead, and I don’t even look at her! The behavior gets absolutely no reward… or even a response, lest that is enough to encourage her.

However, I didn’t feel that it was fair to ignore undesirable behavior without rewarding her for desirable behavior, so I also drastically increased the amount of reinforcement that Maisy gets for appropriate behaviors (such as sitting quietly, lying down, looking at me, etc.), both in general and especially in the face of a trigger.

So far, it’s going fairly well. She’s confused, and doesn’t understand why she is no longer getting a reward for something that’s always paid off in the past. The frequency of the behavior hasn’t lessened at all, but I think maybe the intensity has. I think this will work, though I must confess that I'm worried about the likelihood of an extinction burst. With my luck, it will happen at one of the two trials I am planning to attend with her in the next month.

Still, it will be nice to be smarter than my dog, even if it is just for a fleeting moment.


Laura, Lance, and Vito said...

huh. that is a really interesting conclusion and I honestly don't know what I think. From everything I've read they have all said that classical conditioning overrides operant conditioning. So even if your dog was barking/lunging/reacting you are still going to change the emotional response of your dog by pairing the other dog with cookies. In theory once the emotion starts to change so would your dog's behavioral. And I would expect there to be some middle point where the dog is still a bit nervous but doesn't react right away and you jackpot that. Plus everything should be done as much as possible below threshold so the dog is getting more treats from not reacting.
Of course in real life things go differently and you can't always be below threshold.

So I'm not saying that you are wrong about Maisy's new behavior chain, but it's hard for me to grasp that from what I've read. I would be really interested to see what happens in the next few weeks!

But assuming that Maisy does have this new chain, you are right in that even a glance in her direction could reinforce it. I think that is why I would still try and focus on the "yes" part of what you want her to do- sit, come fore, etc. and try to cue that before she reacts. of course if she reacts and then you cue a come fore that cue could reinforce the reaction. ahh, dog training is confusing!

Anyways I'm really just writing the thoughts that are popping into my head right now. I'm not really trying to disagree with your conclusion :)

Crystal said...

I agree that this is not exactly the expected/normal course for a reactive dog! Maisy has been nothing if not unusual. :) And, based on what I've seen with Maisy, yes: The classical conditioning has worked. She definitely isn't reacting emotionally most of the time. (There are a few moments where I can tell she is more stressed, but interestingly, in those moments, she chooses to play Look At That instead of lunging/growling/barking.)

I think I wrote that I heavily reinforce/jackpot any time she chooses a "non-reactive" response to a known trigger. If she sees a big dog and starts forward but doesn't actually lunge or make any noise, I treat the heck out of her. If she plays Look At That, it's a big party. In other words, I completely agree that it's necessary to focus on the yes part. The reason I chose to completely ignore the behavior is because I didn't want to build in another step to the behavior chain... can you imagine her lunging, barking, and then doing a perfect front? Because I really believe that's what would happen, and I'm not good at being random enough at preventing it.

Here's the interesting/encouraging initial finding:

In my CU class this week, I ignored all reactive behavior. When I did, I waited for her to offer something desirable- lying down, sitting in heel, etc. I treated that. She was able to offer that desirable behavior fairly quickly- within 10 seconds or so of her outburst. And, she was incredibly relaxed when she took the reward- no shark teeth.

This is a huge difference from the way it usually worked: lunge/bark/growl, turn back, get treat, and frantically take treat with shark teeth. It would take her 20-30 seconds to relax and take the treat calmly again.

I'm not sure if I explained that well, but basically, it seems like being rewarded for the reactive behavior actually made her more aroused. Not being rewarded for her seems to defuse or derail the emotions.

I'm really not sure what will happen going forward. This could be really awesome, or I might conclude that it wasn't as helpful as I'd hoped. The hard part is that I'll need to be able to tell if worsening behavior is just an extinction burst, or if it's actually worsening....