I have thought for some time that Lightning’s continued tendency to play keep-away instead of bringing me the retrieval article has to mean that Lightning actually prefers his keep-away game to retrieving.

But that conclusion raised questions for me despite seeming apparent in his behavior: Why was this true for Lightning but had not been true for Lumi and Laddie, the two Goldens I trained previously, who learned a reliable retrieve under field conditions earlier in their training than Lightning has? Why doesn’t preferring keep-away apply to playing with the tennis ball in the house, where Lightning always brings me the ball and has pretty much never played keep-away? Perhaps most important, why would a preference for the keep-away game occur in a dog like Lightning, bred specifically for success in field trials, where keep-away during a competition would be disastrous?

And also, why am I making progress — slow but unmistakeable — in training him to complete his retrieve rather than switch to keep-away? That is, if he prefers retrieving, why did he ever play keep-away? And if he prefers keep-away, why is he learning, albeit slowly, not to do it?

Meanwhile, I recently began thinking about a scenario from game theory called the Prisoner’s Dilemma. You can read about on the web if you like, but basically it creates a situation where each of two parties will take one action, which they reason to be in their own best interest, if they are not cooperating, and an opposite action, which objectively actually is in their own best interest, if they are cooperating.

An example of the concept applied to international relations is disarmament: If one party does not know whether the other party will disarm, then it would arguably be against their best interest to unilaterally disarm and risk subjugation or annihilation. Therefore both parties come to the same conclusion not to disarm and instead engage in an expensive and dangerous arms race. But if both parties believed the other would cooperate, the best solution would be for both to disarm, saving money for each of them and reducing the risk of war.

Other examples of the Prisoner’s Dilemma come up in other fields. And an extended case of it comes up in behaviorism. It turns out that with an Iterative Prisoner’s Dilemma, if the number of iterations is not predetermined, both subjects will eventually learn via operant conditioning to select the optimum solution and begin cooperating.

So following that line of thought, it occurred to me that Lightning’s keep-away games look like a classical Prisoner’s Dilemma. That is, he plays keep-away even though, due to his breeding, retrieving should be preferred. Why? Because at least with keep-away, he gets to keep his toy. Sure, if he delivers it to me and I throw it for him again, that would be more fun. But only if I throw it again.

As in the classical Prisoner’s Dilemma, not being telepathic, he doesn’t have a way of knowing what I’ll do — that is, whether I’ll throw it again — so he takes the suboptimal solution of playing keep-away, in which he deems he is still better off than the worse outcome of giving up the toy and not getting it back again.

But that does not mean that he actually prefers keep-away to retrieving. It just means that he doesn’t know whether I’ll throw the article again if he gives it up.

Further, as Lightning and I work through our version of the Iterative Prisoner’s Dilemma, he is gradually learning via operant conditioning that I almost invariably will throw the article again, or offer some other enjoyable activity, a preferable outcome to being stuck playing keep-away alone.

I think this reasoning explains Lightning’s behavior in a way I’ve never understood before. And I think it also suggests that the key ingredient to eventual success will be patience on my part.

We haven’t done enough iterations yet. I just need to continue the training, and can reasonably expect that eventually, through operant conditioning, Lightning’s keep-away behavior will extinguish from lack of reinforcement.