Abstract
The prefrontal cortex has long been implicated as playing a special role in coordinating and integrating plans of action based on combining sensory signals from the environment and viscera with motivational signals from the organism; this is why it was called “the executive of the brain”. However, the baffling variety of symptoms arising from lesions to different prefrontal sub-regions has led some researchers to doubt that prefrontal functions can be fit into a central “executive” role. In fact, some researchers even hinted that the roles of different sub-regions are distinct enough that the prefrontal cortex should not be thought of as a unit.Recent imagery and lesion data have led us not to reject but to reformulate the executive concept. The data indicates that the brain has an executive system that is hierarchical and yet distributed rather than strictly localized. We argue that the various sub-regions of the prefrontal cortex perform a unified set of interrelated roles, and that these interrelations need to be studied through a neural network model incorporating the frontal lobes and other regions (e.g., basal ganglia, thalamus, amygdala, hippocampus, cingulate cortex, cerebellum, temporal cortex, and parietal cortex). The latest imaging data does indicate that there is a hierarchy towards the frontal pole [1]; this will be included in our model.From the viewpoint of neural network modeling via non-linear dynamical systems, it is useful to break down the overall executive function (EF) into smaller components that repeatedly occur in different tasks and in different combinations. The tasks whose models will be described in Section 3 suggest three generic components that commonly occur, all related to aspects of prefrontal function: (1) establishing links between working memory representations, which could represent sensory stimuli, potential motor actions, et cetera;(2) creating, learning, and deciding among high-level schemata that embody repeatable, but often flexible action sequences;(3) incorporating affective evaluations of sensory events or potential motor plans and using these evaluations to guide actions.The network theory we develop proposes to describe how the components of EF could arise from frontal system neuroanatomy. The phrase “frontal system” refers to the actual prefrontal cortex in conjunction with the areas of thalamus and basal ganglia that are closely connected with prefrontal cortex, and those parts of the limbic system relevant to affective aspects of EF.The model relies extensively on various recurrent loops within the cerebral cortex, between cortex and thalamus, and between cortex, basal ganglia, and thalamus, the latter analogous to the five loops of [2]. These are summarized in the network architecture called ACTION ([3-7]; see Figure 1). The dynamics of this class of network can be described as flow toward attractors that arise from either corticocortical or corticothalamic recurrence. The basal ganglia, via selective disinhibition, help to determine which of the corticothalamic attractors is approached.The details of our general model will be provided in Section 4, preceded by a review of frontal system anatomy in Section 2 and a summary in Section 3 of some models of specific tasks (sequence learning, & WCST) that have already been simulated and that are special cases of our general theory. Further extensions are considered in section 5, and an overall conclusion reached in the last section.