Representations in the central nervous system are population encoded. Understanding the computational processes subserved by pools of cortical and connected subcortical neurons constitutes one of the major challenges facing systems neuroscience. The science of parallel distributed processing (PDP) combines neural plausibility, theoretical coherence, and a demonstrated ability to account for an enormous range of phenomena in normal and damaged human brains. PDP features have now been demonstrated in mice and Hydra. I here discuss a recently introduced granular PDP model of the basal ganglia (BG) that takes into account the fundamental anatomic and neurophysiologic features of the component structures and logically accounts for the effects of low dopamine levels as observed in Parkinson’s disease (PD). Research on lamprey and Drosophila suggests that the essential computational function of the sensorimotor BG (smBG) is reduction of a high dimensionality input space (sensory, motor, and internal drives) to a low dimensionality output space comprised of a limited portfolio of mutually compatible behaviors. Optimization is achieved in the process of iterative settling into a constellation of attractor states. An evolutionary perspective suggests that, for much of the history of complex nervous systems, dating back to arthropod precursors, the smBG, in conjunction with PDP, has provided the basis for the most fundamental of computational functions, reactive intention: the automatic translation of available afferent information into an optimal behavioral response. Experience with pallidotomy for treatment of PD suggests that, in humans, the smBG has become largely anachronistic, its function superseded by cortical mechanisms.

Parallel distributed processing, Population encoding, Attractor, Basal ganglia, Brain evolution, Lamprey, Drosophila