Monday, March 03, 2014

Another Round of Conditioning, Please

The Consequences of Consequences | The Science Dog

Thanks to my friend Gina Keesling for finding this article.  It's a clear explanation of the difference between positive and negative reinforcement, and reports the findings of a study that has some validity in terms of the model used that can be easily replicated for validation.

To summarize, this was a study with dogs that sought to discover which of the modalities--positive or negative reinforcement--was the more effective and whether either (or neither) created stress in the doggie subjects.  The clear conclusion was that negative reinforcement as the major form of training can cause stress in the subjects.  Stress was measured by averted eyes, lip-licking, stress yawns, etc.  There are photos in the article (linked above), so go look at them for clarification.

Dolly responding to the "smile" cue (for which she did not win
those ribbons) from Jess

What I like about this particular study is that the experimenter also included positive and negative punishment in the mix.  The term punishment has become taboo because it smacks of, well, smacking and the like.  In reality, punishment can be something as simple as removing access to food treats in order to extinguish an unwanted behavior.  Don't like your dog jumping on you when you sit down to watch Wheel of Fortune?  Put your knee in her way.  That's "positive punishment"--the addition of something she doesn't like.  Gina and I both opt for the "dreaded Eh!" sound as a precursor cue that means, "Knock it off or I'll stick my knee in your face!"  Then when that knee suddenly appears, it's not as if we didn't warn the animal.

Negative punishment is where the rubber meets the boot heel.  That's the fine line that is sometimes crossed from  "I'm going to take your cookie away" to "I'm going to beat you over the head with a stick".  That's the line that needs to be made firm.
Years later, Dolly's "smile" behavior is 100%
accurate without the visible cue (or any cue,
or any desire on my part for the behavior).

We're moving on to horses, because that's where this discussion actually came up.  There are many levels of negative punishment that are useful in inhibiting unpleasant learned behaviors in our equine partners.   One of my geldings developed the unpleasant habit of leaning into me as I walked him on the lead.  I don't know why.  My guess was that he hoped to direct me instead of my directing him, but I couldn't get him to confess to that.  After lots of attempts to use positive reinforcement for the "off me" cue ("Yay!  You're off me!"...cookie) and pretty much no change in his behavior, I moved to negative reinforcement.  "You want to lean into me?  Let's just circle to the left twice and back up five paces and stop when you aren't leaning."  That was fun, but it gave him a whole new slew of behaviors with which to annoy me.  His leaning turned into moving his butt away from me and rapidly circling to the left.

Leo takes his rewards in whatever form they come.
Finally, rather than sell him as a "leaner" and buy a different horse, I remembered a stupid human trick as described in Walter Farley's The Black Stallion.  The main character, tired of being bitten by the horse, put a hot potato under his sleeve and made it easy for the horse to bite the potato instead of his arm.  One-trial learning occurred.  Of course, that's a novel, not reality.  But in theory it made sense. So I started carrying a crop when I lead my boy.  I didn't hit him with it.  I held it straight out to the side so he couldn't close the gap without running into it.  This translated at shows into my elbow sticking slightly out to the side, but at home the crop did wonders.  In no time he'd figured out where he needed to be, and behavior modification was achieved.

Regular readers are probably tired of hearing me say that using only positive reinforcement is lovely, but it makes for a very loooooong learning curve as the animal can exhibit  hundreds of behaviors before he accidentally hits on the one you want and you reward him for it.  Then there's the process of narrowing down the options because even though you think you rewarded him for, say, taking a step to the right, at the same time he did that he was also looking at a leaf, twitching his nose, swishing his tail, thinking about dinner, and so on.  He has no clue which of those things you rewarded unless you used a clicker to pinpoint the moment.  Even then there's a small curve, but it's easily mastered.  Negative reinforcement combined with positive and negative punishment in small, reasonable doses serves to clarifies the requested behavior for a critter who doesn't share our language.

Getting to positive is the goal.  When there are 100% replications of the desired behavior, then there are no stressful negatives involved, and everyone goes home a winner.  Cutting to the chase using the most efficient methods makes that journey a happy one.

No comments: