As I've reviewed previously, Fuster’s Network Memory Theory proposes that hierarchies of cortical neural networks may be formed through experience, and that this arrangement explains the full range of behaviour, from the most basic of movements to complex movements and planning. The development of these hierarchies depend not only on the direct experience of the individual, but also on the 'memory of the species' upon which these may be based: "Memory is formed in the cortex from the bottom up, along ontogenetic gradients, from primary areas to progressively higher areas of association." The purpose of the present paper (ref at end) is to describe the perception-action cycle (by which the hierarchies are integrated with the environment), and a description of what goes on at the upper levels of the cortical hierarchies. I'm considering this latter point to be an important consideration, as an undefined 'head' of the respective hierarchies may lead the theory open to problems of having to explain seemingly all-powerful executive constructs (my opinion, I may be misplaced here).
As may be self evident (though not according to the enactive perception view for example), sensory information is necessary to determine which actions are appropriate at any given moment. These actions would change the state of the environment of some way, which in turn would change the sensory information obtained. This process is thus circular, leading to the term 'The Perception-Action cycle'. In the paper, neuroanatomical and neuroimaging evidence from humans and other primates is presented in support of this view: "there is anatomical evidence of an orderly descent of connections from prefrontal to premotor to motor cortex...", and, "... as Koechlin et al. also show, the activation of progressively lower frontal areas that process the action is cumulative." The evidence then points to a situation where "Automatic and well-rehearsed actions in response to simple stimuli are integrated at low levels of the cycle, in sensory areas of the posterior (perceptual) hierarchy and in motor areas of the frontal (executive) hierarchy." Furthermore, it also supports the prediction of the perception-action cycle that sensory information flows from posterior sensory regions towards frontal controlling (or executive) areas of the hierarchy structure.
So, what happens at the top of the hierarchies? As has been mentioned, the lower levels of the hierarchy are responsible for the relatively simple actions - as one 'rises' in the hierarchy, the actions and perceptual stimuli represented would be more complex, and more abstract from the real world grounding of representations lower down. Cooperation, and eventually integration, of the hierarchies would thus become more important at these higher levels. The cortical areas postulated to be at the top are the prefrontal cortex (particularly the rostral area thereof), and areas of the higher sensory association cortex. The Koechlin et al study mentioned previously also showed, using fMRI analysis, the descending activation down the hierarchies, supporting the theorised order of processing in the frontal hierarchy (the upper levels influencing the functioning of the lower levels).
Given that this difference exists as one rises up the hierarchy, how are actions decided when there may in some instances be competing options from either ends of the hierarchy. Fuster proposes that the next action in a sequence of actions is determined under two influences: the current sensory situation, and "the processing of the global aspects of the sequence in upper frontal areas...". As mentioned in the previous paragraph, the upper layers of both hierarchies 'inform' those below - before this may occur, the information in both hierarchies is integrated with previous information. In this way, the prefrontal cortex integrates temporal associations which are stored in the networks of the perceptual and executive cortex.
As a final comment, a note is made on serial and parallel processing in the hierarchies. "Contrary to common misconception, nowhere in an ascending or descending cortical hierarchy does processing need to be exclusively serial. In part because of the dependence on feedback throughout, the processing in the perception-action cycle takes place not only in series but also in parallel."
J.M. Fuster (2004), "Upper processing stages of the perception-action cycle", Trends in Cognitive Sciences, vol. 8, no. 4, pp143-145
Koechlin et al (2003), "The architecture of cognitive control in the human prefrontal cortex", Science, 302, pp1181-1185