Drive Training and the Missing Mechanism Behind Premack’s Principle
Why Premack’s Works When It Works, Why It Doesn’t When It Doesn’t
Cats, Monkeys and Guinea Pigs
In another post from a few years ago,I talked about the predatory sequence in wolves—the search, the eye stalk, the chase, the grab bite and the kill bite—and how this sequence plays out in various breeds of dogs and the behavioral tendencies each breed is best known for. I also discussed how almost all obedience behaviors—except the sit—were originally based on the behaviors found in this sequence. I believe this is why drive training in general, and Natural Dog Training in particular, is the most effective and comprehensive model available for the training of pet dogs. Nothing else comes close.
In fact the closest that trainers using “positive” methods comes is when they apply what’s called Premack’s Principle which basically posits that a more probable (or high-frequency) behavior can be used to condition a less probable behavior. An example often given is that of the parent telling the child, “If you eat your spinach (a less probable behavior) you can have some ice cream (a more probable behavior).”
Since most obedience behaviors in dogs are based on the motor patterns found in the way wolves hunt large prey as a cohesive social unit, those sets of predatory behaviors are more probable in dogs than they would be in, say, cats, monkeys and guinea pigs.
In 2013 Kevin Behan, who created the Natural Dog Training model of behavior, wrote on his blog: “Basically the Premack principle is a thin slice of the Drive principle. However, just like the ‘high value’ reward concept in behaviorism, what makes a behavior higher or lower in frequency is always its relation to the prey drive and the Premack principle fails to make this distinction and this will have consequences in training. Making prey (chasing balls, sticks, digging, chasing deer, squirrels, going for car rides, etc., etc,.) is the most frequent behavior a dog can perform so due to this, the Premack Principle is somewhat related to the Drive principle, but Drive is a more precise understanding of the learning phenomenon. It elucidates an underlying dynamic while modern learning theory, such as the Premack principle, is exclusively focused on reinforcement.”
Yet as we shall see, Premack’s definition of reinforcement doesn’t include an underlying physical or emotional mechanism. As Gertrude Stein famously said of Oakland, California, “The trouble with Oakland is that when you get there, there isn't any there there.” In like manner, when you parse out the mechanism behind why Premack’s Principle works, there isn’t any there there either.
Six Conditions In Search of a Mechanism
Nevertheless, Premack’s predicts six conditions, all of which have been supported by evidence, as reported below on Wikipedia.
1) Reinforcement is a relative property. Responses A, B, C have a descending rank order of probability. A will therefore reinforce both B and C. C will reinforce neither. This suggests that reinforcement is an absolute property. However, B corrects this view. B will reinforce C, but not A. B is both a reinforcer and not a reinforcer. Reinforcement is therefore a relative property.
Okay, so reinforcement is a relative property. But what ‘s the underlying mechanism causing some things to reinforce behaviors and prevents other things from doing so?
2) Reinforcement is a reversible property. When drinking is more probable than running, drinking reinforces running. When the probabilities are reversed, running reinforces drinking.
Here we run into another wall. Reinforcement is dependent on probabilities. What causes those probabilities? We’re not told.
3) Historically, consummatory responses, eating and drinking, have served exclusively as reinforcers, but consummatory responses are, like any other response, subject to reinforcement.
Here we’re about to get close to something tangible: hunger and thirst act as reinforcers for learning. Yet they’re also described as being subject to reinforcement. In other words (if I’m understanding this correctly), food and water act as reinforcers for other behaviors but the acts of eating or drinking are themselves dependent on some form of reinforcement.
4) Reinforcement and punishment, traditionally contrasted as opposites, are in fact equivalent except for sign. If response A leads contingently to B, and B is more probable than A, A will increase in frequency (reinforcement); conversely, if A leads contingently to B, and B is less probable than A, A will decrease in frequency (punishment). The major contrast is not between reward and punishment; but between reward and punishment as contrasted with freedom. Freedom is the condition in which stimuli are freely (not contingently) available to an individual.
I find this section very interesting, yet there’s still no “there there.” By this I mean that rewards and punishments are not defined here by their physical or emotional qualities, if any, i.e., things that can be felt, touched, observed and measured. They’re defined by their contingencies: something liable to happen as an adjunct to or result of something else possibly happening. Only freedom seems to have some measure of reality to it, yet it’s still defined in contrast to things that only exist as probabilities, having no real substantive physical or emotional qualities.
5) When motorized running is more probable than lever pressing but less probable than drinking, then running reinforces lever pressing and punishes drinking. In other words, the same response can be both a reinforcer and a punisher - at the same time and for the same individual.
Again, these things are being defined purely through probabilities, not any substantive physical properties of things or feelings.
6) The equivalence of reinforcement and punishment is further suggested in this interesting fact: rats are either sensitive to both reinforcement and punishment, or insensitive to both; they are never sensitive to one but insensitive to the other.
At last there’s a feeling involved. Yet one has to wonder, how can a rat be sensitive or insensitive to things – reinforcers and punishers – that don’t have a substantive nature?
Notice that what is viewed as a reinforcer in Premack's Principle is not a stimulus—meaning something physical like food or water—but a behavior, like running or drinking
I find this all very strange.
“Well,” Premack supporters might say, “the point is that more probable behaviors can be used to condition less probable behaviors. There’s no dispute about that.”
But there is, actually.
Violating Premack’s
From the website of Indiana University: “In 1974 [William] Timberlake and [James] Allison conducted their own experiments to test Premack’s Principle and found that it did not always hold true. Their results led them to propose a new formulation which they called response deprivation: a less probable behavior will serve to reinforce a more probable one if by engaging in the baseline amount of the probable behavior, a subject is thereby deprived of the baseline amount of the less probable behavior.
“As an example, let's say that during paired baseline, a rat spent 20% of its time running in the wheel and 10% of its time drinking. We now arrange a contingency phase in which every four minutes spent running earns the rat one minute of drinking (a 4 to 1 ratio). If the rat now spends the same amount of session time running as it did in the baseline phase (20%), it can spend only one fourth of that time (5%) drinking, or only half of the baseline amount of drinking (10%). This contingency thus deprives the rat of drinking, compared to the baseline amount. According to the response deprivation analysis, the amount of running should increase over baseline, in order to restore drinking to baseline levels. In other words, drinking will reinforce running.
“Note that in this example, Premack’s would predict that drinking would not reinforce running, as during baseline, drinking was the less probable behavior. Experimental tests showed that the response deprivation prediction was correct and not Premack's.
“Similarly, it is possible to arrange a contingency such that, by engaging in the baseline amount of the less probable behavior, the rat can obtain at least the baseline amount of the more probable behavior. Response deprivation predicts that, under this contingency, the more probable contingent behavior will not reinforce the less probable instrumental behavior. This prediction was also confirmed, again violating Premack's Principle.”
Yet even though it has been proven that Premack's doesn't always hold, very few trainers have ever read Timberlake and Allison's work, or have had occasion to apply it. Why not? Because, as Kevin Behan says, Premack’s is a slice of the drive principle, and the drive principle is the most important and effective aspect of dog training.
Chasing Squirrels
Back in the early 1990s, my Dalmatian Freddie was fond of chasing squirrels in Central Park. He was not what I would call an “instant chaser” but more like a “patient stalker.” He’d see a squirrel, go into a stalking stance then wait patiently until the squirrel took off running. Then he’d run full-blast, chasing the little critter until it ran up the nearest tree.
Then, one day, Freddie actually caught a squirrel. When I saw him come up over a low hill, with the poor creature in his mouth my heart sank. Fortunately, just as Freddie was not an instant chaser he was also not an “instant biter.” So, as soon as I saw him I gave him the “Out!” command. He immediately dropped the squirrel and it ran up the nearest tree.
No harm, no foul.
Still, I thought that wasn’t a very enjoyable experience for the poor squirrel. Then I thought, I have to do something to keep this sort of thing from happening again. So I resolved to find a way to do that. So, the next morning, I didn’t feed Freddie his breakfast. Instead I took along a plastic baggie full of juicy chicken bits. We got to the park and I let Freddie do his usual thing, sniffing around, etc. Then, when he caught sight of a squirrel and froze in his stalking stance, I took out a juicy piece of chicken and waved it under his nose.
He not only ignored it but, since my hand was blocking his view, he moved his head slightly away so he could keep his eye on the squirrel. Every time I moved the tidbit, Freddie craned his neck around to get a better view.
Huh, I thought. This doesn’t seem to be working.
Then, since his mouth was open slightly, I put a piece of chicken right onto his tongue.
Surely this has got to work, I thought.
Nope. He just let it sit there for a moment, then spit it out!
Now I was really perplexed. There seemed to be no way to solve this problem using food rewards. This realization was not just perplexing, it was depressing. The only other solution seemed to be to keep him on the leash at all times. But even that wouldn’t solve the problem because, leash or no leash, once Freddie saw a squirrel he would not move an inch until the squirrel started running for the nearest tree.
I meditated on the problem for a while, then frustrated and angry that I couldn’t think of a solution, I made a threatening move toward the squirrel, hoping to get it to run up the nearest tree so I could at least continue my walk with Freddie. Sure enough, the squirrel ran toward the nearest tree. Freddie raced over to try to intercept him but the squirrel was faster So Fred ran in circles around the tree, barking furiously at the squirrel.
Then I remembered that like wolves, dogs are social predators. If I could treat this as a hunting exercise, where the human handler controls a hunting dog’s behavior, I might be able to get somewhere. For instance, I thought, if I could put Freddie in a down/stay, that might solve the problem. But even then, he would probably hold the down/stay until the squirrel started running, so I was still at an impasse because my goal was to not just to get him to stop chasing squirrels, but to keep him from stalking them as well.
Hunting Partners
Then it hit me. What if I did what I’d just done? What if I started stalking the squirrel first and pretended that Freddie and I were hunting squirrels together? Unless they were moving, I was usually better able to see them. So what if I saw one before he did, said something to Freddie to make him feel we were on the same wavelength, and then started chasing the squirrel myself?
So, I put the chicken in my pocket, and later, while Freddie was sniffing around, I spotted another squirrel, one that he hadn’t seen yet himself.
In a hushed, highly emotionally charged voice I whispered, “There he is!” and began stalking the squirrel myself.
Freddie eventually picked up on my mood, and when he did, he saw the squirrel too, and dropped into his stalking stance. We were now hunting together. Fred didn’t know it yet, but I was now in control of the game.
I did this for a few days then added a new twist. We’d stalk a squirrel together but at some point, I’d make a quick move toward the squirrel, motivating it to run up the nearest tree. This always set Freddie racing off after the the little critter. While he did that, I'd pick up a stick, hoot excitedly and run away, waving it for Freddie to see.
Freddie would then be forced to choose between chasing the squirrel and ending up circling the tree to no avail, or chasing me and the stick. In the beginning he always went immediately for the squirrel. But the squirrels always went up a tree, leaving Freddie with nothing to sink his teeth into. That’s the critical thing here.
Meanwhile, I was still enticing him with a stick. Once he started to come toward me, I’d run away, shouting, “Freddie, come!” (even though he was already in the act of running toward me). Then, once he got to me, I’d invite him to jump and play tug-of-war. I either let him win, or, if he lost his grip, I immediately threw the stick for him to chase, which he did with the same intensity he had when chasing the squirrel. Once the stick was in Freddie’s mouth and he was able to lie down in the grass and crunch down on it with his jaws and “kill” it, he was truly satisfied. He never got that satisfaction from chasing squirrels because even when he caught one he was too tender-hearted to bite it.
After just a few weeks of following these steps, whenever Freddie saw a squirrel, all I had to do was pick up a stick, then whistle, or say, “Freddie, come!” and he’d immediately turn and run back to me for a game of tug or fetch.
Here’s how drive training—or my understanding of it—differs from Premack’s. When Freddie saw a squirrel he became filled with an emotional charge. He was so charged up in fact that nothing could get his attention away from his intended prey, not even a juicy piece of chicken sitting on his tongue! By pretending to immerse myself in the same emotions he was feeling, I created a dynamic, magnetic charge between us. Then, by getting him to jump up on me and play tug-of-war, I decreased his resistance to my position as a vertical being and gave him the satisfaction of crunching something with his teeth.
Meanwhile, this is not easily explainable through learning theory, where reinforcers are defined as anything either added (positive reinforcement) or subtracted (negative reinforcement) increases the statistical likelihood that a behavior will be repeated, and punishers are defined as anything added or subtracted that increases the statistical likelihood that a behavior won’t be repeated.
So when I said to Freddie—“There he is!”—it increased the statistical likelihood that Freddie would look to find the squirrel, meaning that behavior was reinforced. When I got the squirrel to run away, that behavior increased the likelihood that he would chase the squirrel, meaning that behavior was reinforced. So instead of actually teaching Freddie not to chase squirrels I was increasing not only the likelihood that he would chase them, I was also increasing the intensity of his response.
Then, when I picked up the stick waved it and ran away, while Freddie was circling the tree trunk, I increased the likelihood that he would come chase me. And when I invited him to jump up and play tug, I increased the likelihood that he would repeat those behaviors.
Since doing all this caused Freddie to stop chasing squirrels altogether—in other words that set of behaviors was extinguished—and since behaviors can only be extinguished through positive or negative punishment, where in this chain was Fred being punished?
This is where Premack’s comes into play. I was using what was a more probable behavior at the time—chasing squirrels—to condition a set of less probable ones—chasing me, jumping up on me, and playing tug-of-war. Where this differs from Premack’s is that were actual, physical and emotional mechanisms at play, and those mechanisms were directly related to the fact that dogs—like wolves—have a constant need to express their predatory instincts and emotions as part of a group dynamic. In other words, I wasn’t “punishing” Freddie’s behavior of chasing squirrels, nor was I conditioning him to engage in a “less probable” set of behaviors. I was giving him a deeper, more emotionally satisfying outlet for his predatory instincts, in particular, the urge to bite.
That’s the drive training difference. It operates on a level all dogs and dog owners can easily understand: feelings and emotions instead of contingencies and probabilities.
Lee Charles Kelley
“Life Is an Adventure—Where Will Your Dog Take You?”
