Positive Reinforcement vs. Drive Training

Which Training Approach Is Best?

In certain training circles it’s widely believed that learning theory is the only truly scientific and, therefore, the only correct approach for training dogs.

Is this true?

Not exactly.

To be fair, learning theory (also known as behavioral science or behavior analysis) is much more scientific than dominance theory, especially as it’s applied to dog training. There is plenty of scientific evidence showing that dogs and wolves form dominance hierarchies, but none showing that such hierarchies can cross species boundaries.

Still, while dominance training is based mostly on fantasy, not science, positive training is not based on hard science. There are no underlying scientific principles behind how learning theory works in the way that there are in physics and chemistry, where natural phenomena are explained through specific scientific laws. Learning theory is explained only through statistics. Don’t get me wrong. Statistics are important. But they don’t describe the how and why of behavioral changes in humans and animals, only the when and how often.

There are two other problems. And they’re huge.

First, modern, 21^st Century research has shown that behavioral changes don’t take place through associative learning, the theoretical process where a human or animal associates a behavior with a reward and thus wants to repeat that behavior to get another reward. etc. In fact, B. F. Skinner—who created this model of learning—said that the only way to ensure that learning will stick is to provide a series of initial rewards, then start withholding expected rewards using what’s called a variable reinforcement ratio. And unless you have an advanced degree, it can be very difficult to do this with any accuracy.

So that’s the first problem. 

The second is that in actually studying the effects that take place in the brains of animals during the learning process, scientists have found a puzzling paradox: dopamine— often referred to as one of the brain’s reward chemicals—is not released in connection with an external reward, it’s released in the absence of an expected one. Other research shows that dopamine is not really a reward chemical at all. It’s more like a salience detector, meaning that it helps humans and animals pay close attention to changing patterns in their environment. The upshot is that learning doesn’t take place through forming an association between a behavior and a reward but rather through a mostly unconscious process called pattern recognition.

I’ve found that an easy way to determine if this is true is to teach a puppy to sit, but do so, out of sequence, in the following manner: Show the puppy a treat. As he tries to grab it, move it around in such a way that at some point the pup sits on his own. As soon as he sits give him the treat. Then—after he’s already eating his “reward”—wait a moment, then say “Sit!” in a happy voice.

Do this three or four times. Then, show him the treat again and say “Sit!” without waiting for him to exhibit the behavior. Wait a half-a-second or so, and the pup will automatically sit.

Go through this same process again later, in a slightly different environment, or at a different time of day. And once you’ve done it several times, under slightly different circumstances, all you have to do is show the pup a treat, say “Sit!” and the pup will automatically sit.

If learning takes place through associating a behavior with an external reward, the pup wouldn’t learn to sit through this “backwards” way of doing things. Or at the very least, it would take much longer for the pup to learn the behavior and repeat it in a reliable way.

Of course, the sit is among the easiest and most basic behaviors we can teach a dog. And some readers may not see the distinction I’m making between associative learning and pattern recognition. So let’s look at the very complex behaviors exhibited by some working dogs.

Border collies operate almost entirely on pattern recognition. They have to take into account the movement of the flock as a whole, the behaviors of the individual sheep, the changing nature of the terrain, the signals sent to them by the shepherd, and a whole host of other factors. This is one reason why border collies are among the “smartest” dogs. And it’s all due to pattern recognition. There are very few, if any, external rewards provided when these dogs are learning their trade and none while they’re working.

It’s also well known that you can’t train police dogs, military dogs, bomb-sniffing dogs, drug detection dogs, and search-and-rescue dogs using conditioning techniques. That is, you can, you just won’t get a good result if you try to do so.

Why?

It’s well known that detection dogs will give false alerts in order to get a food reward. Meanwhile, police dogs are never trained through rewards but through games like tug-of-war that stimulate and satisfy the dog’s urge to bite, a form of drive training that is quite different from reward-based techniques.

As for search-and-rescue dogs—who often need to be able to navigate the rubble and detritus of an urban disaster site—one training technique used is a scenario where the dogs are urged to climb up or down fire escapes while their trainers throw pots and pans in their direction. Talk about the opposite of positive reinforcement! Yet doing this actually makes the dogs perform better.

There many other examples where the behavioral science model and drive training diverge. 

The point is that when dogs are trained through elements of the wolf’s prey drive—the search, the chase, the grab-bite, and the kill-bite—they learn and operate through a completely different set of principles than those espoused by reward-based trainers. And not only that, but the kind of learning and obedience that takes place with drive training is far superior to what you get with reward-based training. 

Lee Charles Kelley

“Life Is an Adventure—Where Will Your Dog Take You?”