The rational internal competition and selection hypothesis

One way to envision how behavior occurs is to imagine that the brain has distinct icons for distinct behaviors, much as there are distinct desktop icons for distinct computer programs, and the brain activates a programmed motor behavior by essentially double-clicking on the relevant icon. There are no icons, however. Instead, actions begins as memories, which are patterns that neurons tend to form, e.g., as attractors in a dynamic system, based on previous experience. As the effects of a particular formation propagate across the brain and body, the result is something that a human observer would classify as an action—or a perception.

Since there are no predetermined correct responses to given, stereotypical situations, there is no automatic way to determine what behavior to construct in a situation. And since the environment contributes to the determination of what is constructed, and because the internal environment is unreliable, the effects of any particular neural assembly are not given. The same assembly will produce different effects depending on both internal and external conditions. Thus, even if it is clear what the intended behavior should be, it is not clear which neural assembly will produce that behavior; similarly, even if it is clear what behavior will be produced by a neural assembly, it is not clear which behavior is optimal.

Thus, the challenge of constructing the optimal behavior is much more difficult that it first appears. Nothing is given, neither the optimal behavior nor the behavior that can be expected of an internal assembly. How, then, is optimal behavior determined and produced?

There is growing evidence that behavior is constructed internally by a process of competition and selection. Rather than starting from one neural assembly and hoping that it yeilds the optimal behavior in conjunction with the internal and external milieus, multiple neural assemblies are constructed in parallel, each predicting different sensory data from which the behavioral consequences of the assembly are inferred. The different assemblies then compete against each other, with the predictions of the winner being faciliated across the brain and body while the predictions of the losers are inhibited.

The advantage of the method of constructing multiple options, having them compete, and selecting the winner of the competition to be implemented is that instead of having to determine a priori the right thing to do, the brain can compare several options to see which one is best.

However, there are so many possible behaviors that could be implemented in a given scenario—and so many competing views on what about the scenario is actually given—that even constructing a dozen different competing neural assemblies doesn’t give you good odds of behaving optimally, just like buying a dozen lottery tickets doesn’t mean you’ll win the lottery. Unless you can construct an insane number of competing options, you need a smarter method—you need more information.

Competition is the way to get that information. In this hypothesis, a neural assembly is facilitated across the brain and body by convincing other neurons and cells to join it, which it can only do by convincing it that the model it offers is better than the model offered by the other neural assemblies. Because each cell is an agent, it will only join a neural assembly if it expects to be better off by doing so, factoring it its own personal information or perspective. The neural assembly will therefore adjust the predictions it’s putting out to recruit as many cells as possible, therefore relying on the behavior of these other cells to form the model most congruent with their information. Thus, competition acts as a discovery process, discovering which neural assembly is the best according to the information that is decentralized among the individual cells. Gathering all of this information by adjusting the model is hard, but it is incentivized by the competition process—the assemblies want to be facilitated and do not want to be inhibited.

This process is rational in the economic sense if the competition between the neural assemblies produces an ordering between them that is complete and transitive, meaning that you can definitely say which assembly came first in the competition, which came second, which came third, etc., and if the first-ranked neural assembly, the one that did the best job of convincing the cells of the body that its plan is best, is selected as the plan that is implemented by the organism. If so, then a great many processes that we would not ordinarily call rational, such as processes that occur within our bodies but without our conscious awareness, are in fact rational.