In 1972, Michael Domjam and Nancy Wilson tested an interesting hypothesis: When creating a conditioned association between two events, do the types of events matter?
To help illustrate this question more clearly, let’s take an example: You go out for dinner at a new restaurant. You eat a large meal consisting of salmon, and don’t eat anything else for the remainder of the evening. The next morning you wake up to find yourself violently ill. As a rational human, you are likely to make the assumption that it was the food you ate which made you ill. In other words, you don’t assume it was the particular wallpaper that was chosen for the restaurant, or the uniform worn by the waiter, even though these two things were as equally correlated with the illness as the food you ate. You likely will have at least a temporary aversion to salmon, and no aversion at all to red floral wallpaper, etc. The point being that if the effect is gastro-intestinal illness, then we are more likely to have a conditioned aversion response when the perceived cause is something we ingested, versus something we saw or heard.
While this may seem like common sense, or perhaps simply a function of our human ability to explain the connection via reason and language, Domjam and Wilson set out to investigate further. In their study titled Specificity of Cue Consequence in Aversion Learning in the Rat, they were able to show that when either a specific taste (saccharine) or a sound (buzzer) were paired with illness (via injection with lithium chloride), the rats learned a very strong aversion to the taste, while the group that received the buzzer/illness pairing learned virtually no association at all. Likewise, when they paired the saccharine or buzzer with a shock, the results were the opposite. The group who received the taste/shock pairing learned almost no association, while the buzzer/shock group readily learned the association.
The theoretical implication of this study is that animals will learn selectively, in favor of associations which reflect an underlying causal relationship (Dickenson p.60). Evolution has programmed our learning mechanisms to understand that an external stimuli such as a touch sensation is more likely to be caused by something you can see or hear (such as a predator), while internal discomfort is more likely to be caused by something you ate (such as rotten meat). Clearly animals that have developed such learning preferences have a greater chance of survival.
I recognize that at face value, this may not seem like a groundbreaking concept. However if we take this line of reasoning a bit further, I think it raises some interesting questions, particularly in the realm of determining what constitutes “humane” dog training. 
First, a qualifying statement: At this time I have no idea if any of the following hypotheses have been studied. The following discussion is purely a speculative thought experiment for the purpose shedding light on a different way of approaching the question of what is truly “humane.”…
To begin, knowing that animals employ selective learning in the case of aversion, is it fair to assume that the same selective learning may exist in the case of attraction? If we paired a variety of different stimuli with pleasant outcomes, would the same learning biases occur? For instance, will sight, sound, or smell be associated more readily with a pleasant taste than touch? If we assume the evolutionary perspective on causal relevance, then perhaps the answer would be species specific. Predatory animals rely on sight and sound to find food (prey), but most predators do not rely on touch to find something to eat. However, herbivorous prey animals are likely to use sight, smell, and perhaps even touch (foraging) to find food, and are less likely to rely on sound (although they would rely on sound to avoid predators.) If evolutionary pressures select for learning biases, then it seems appropriate to at least consider these possibilities.
For the purpose of exploring this line of reasoning, let’s assume for the moment that the hypothesis above is correct, and that selective learning applies to both aversion-based associations, as well as attraction-based associations. The next question that I would like to consider is if these biases have any influence on counter-conditioning protocols.
For those unfamiliar with the terminology, counter-conditioning refers to a classical conditioning process in which an attempt is made to alter the emotional association of a particular event or stimulus by pairing it with an event or stimulus which has an opposing emotional effect. A simple example of this would be to take a dog that is afraid of a flag blowing in the wind, and present it with food/treats every time it sees a flag. In theory, if the procedure is executed properly, over time the dog should begin to associate the flag with the presentation of food, and will begin to have a more pleasant emotional response to the sight of a flag.
If we apply our theory of selective learning to counter-conditioning, does it follow that the efficacy of a counter-conditioning procedure would be dependent on the nature of the two events/stimuli/emotions being paired? For example, does fear that is associated with a sight, such as a flag, respond better to food as a counterconditioning agent versus perhaps a toy, or a social reward (affection)?
Additionally, is it the type of emotion that matters most, rather than the type of stimulus? In other words, does it matter if the dog is in fear for her own safety, versus mildly annoyed? Does it matter if the stressor is related to a social anxiety (such as dog on dog aggression), versus a noise phobia (perhaps as the result of something falling and crashing to the ground near the dog)?
To take it one step further, can a negative reinforcer act as a counter-conditioning agent? I have written in the past about research that was conducted in the 1970’s that showed that behaviors that are learned through an escape/avoidance procedure (negative reinforcement) are maintained via a sense of relief and relaxation which occurs upon successfully eliminating or avoiding an aversive. In 1983, Daniel Tortora applied this finding in an interesting way. With the help of Tri-tronics, he designed an electronic collar which automatically produced a tone after the release of the stimulation button/termination of electrical stimulation. There was also a button that could produce the tone only. He used this system to rehabilitate 36 dogs who exhibited what he termed “Avoidance-motivated aggression.” Through the process of obedience training, the dogs learned to associate the tone with the sense of relief/relaxation/safety that they experienced after the stimulation button was released. He also used the tone-only button when the dogs successfully executed a command and avoided any stimulation.
Once the dogs were successfully conditioned to the meaning of the tone, he then put the dogs in a situation which normally would trigger an aggressive response, except he would give one of the previously trained commands, and present the tone to signal safety when they successfully completed the command.
The theory behind this procedure was that the dogs were using aggressive displays to achieve a sense of safety during a perceived threat. Tortora was using the conditioned safety signal (tone), to show the dogs that they could achieve the same sense of safety by behaving obediently and in a pro-social manner.
While I am not certain that this can truly be considered counter-conditioning, the results of his study were very similar to what we would desire from a traditional counter-conditioning procedure. Not only did Tortora report that aggression was successfully eliminated in all 36 dogs in the study (He followed up with each dog’s owner for a period of up to three years after the study concluded, and no dogs required re-training). Most interestingly, it was reported that each dog showed overall improvements in emotional stability, and evidence of an increased sense of “self-efficacy” or courage (Tortora p.176). In fact, Tortora noted that the safety tone, once learned, appeared to act as a conditioned positive reinforcer, thus supporting the theory that negative reinforcement plays a legitimate role as part of a counter-conditioning procedure.
Moreover, Tortora compared the results of his safety training experiment to several other protocols including differential reinforcement of alternative behavior, systematic desensitization, flooding, and punishment. In all cases, safety training was superior to the rest.
If we bring this back to our previous discussion, it appears that there may be some evidence that: A) negative reinforcement can play a role in counter-conditioning (either directly, or indirectly through the development of a unique safety signal which acts as positive reinforcement), B) counter-conditioning procedures can be susceptible to selective learning, and C) in the case of emotions that elicit avoidance-motivated behaviors, a program which involves negative reinforcement may be more effective than positive reinforcement alone for the purposes of counter-conditioning.
If that last statement turns out to be correct, then there are some interesting ethical implications. If we consider that avoidance-motivated behaviors are elicited by unpleasant emotions, then it follows that dogs who are displaying avoidance-motivated behaviors are suffering from diminished emotional welfare. If (1) it turns out that counter-conditioning, facilitated either directly or indirectly by negative reinforcement, is the most effective means of counter-conditioning those unpleasant emotions and producing the greatest overall reduction of stress when compared to other types of reinforcers, and (2) we assert that a training procedure is to be judged as “humane” by assessing its overall effect on emotional welfare and the overall increase or reduction on stress, then it follows that, at least in some cases, a training approach that involves negative reinforcement may be the most humane choice.
Admittedly, this discussion has been largely based on hypothesis. However, I hope at the very least what might be taken from this is that a comparison of various training procedures, particularly in reference to determining what constitutes “humane” dog training, is far more complex than it might initially appear. Oversimplification of the matter will likely lead to a position that does not hold true across the broad range of potential training contexts. In any conversation in which the question “Which training philosophy is better?” is posed, it should always be followed by “Better at what?” The more we define the “what,” the more likely we are to realize that these questions may not be as easy to answer as we once thought.
-Tyler Muto
References:
Domjam, M. & Wilson, N.E. 1972. Specificity of cue to consequence in aversion learning in the rat. Psychonomic Science, 26, 143-5.
Tortora, D.F. (1983). Safety training: The elimination of avoidance motivated aggression in dogs. J Exp Psychol Gen, 112: 176-214.
Dickinson, Anthony. Contemporary Animal Learning Theory. Cambridge: Cambridge UP, 1989. Print.
Denny M.R. (1976). Post aversive relief and relaxation and their implications for behavior therapy. J Behav Ther Exp Psychiatry, 7: 315-321.
Leave a Reply