Complex cerebral activities are likely to be composed of massively repeated simple data processing tasks since the cortical data processing unit, the cortical mini-column, is found throughout the cortex with only minor variations. It has been proposed that one task performed by the cortical mini-column may be to match afferent sensory data to learnt datasets in a process known as automatic association. We hypothesize that basal ganglia circuits, through the relative signal of the nigro-striatal and striato-pallidal pathways, determine the matching threshold for dataset matching within cortical mini-columns. Basal ganglia circuits are in a unique position to use parallel information to modulate the parameters of auto-association to increase the speed of data processing tasks. This hypothesis can explain motor symptoms in Parkinson's disease and also predicts that over and underactivity of basal ganglia circuits (the 'on' and 'off' states) will lead to characteristic errors in sensory data interpretation in all modalities - false negative data recognition when 'off' and false positive data recognition when 'on'. As a preliminary exploration of this hypothesis 16 patients with advanced Parkinson's disease were tested in voice and face recognition when 'off' and 'on'. Each patient exhibited errors in the recognition task according to basal ganglia activity as predicted by our hypothesis. Further experiments to test the hypothesis are proposed.