Understanding of human cognition is hampered by a folk-psychological conception of cognition from the first-person phenomenological perspective. Such a homuncular view has held back our ability to bridge the gap between neurons and behavior. Using biologically plausible computational modeling of frontostriatal circuits, we explore how relatively complex cognitive control tasks can be solved without recourse to a homunculus. Our models build on the more well studied roles of the basal ganglia (BG) in “gating” of motor action selection and reinforcement learning, and suggest that in more anterior anatomical circuits, the BG and reinforcement learning processes guide the gating of cognitive actions, such as when and when not to update prefrontal working memory representations. In so doing, the models have provided a theoretical account of how the brain can learn to learn in complex environments with memory demands, as well as a parsimonious explanation for the interaction of working memory and reinforcement learning. We have further extended these models to investigate hierarchical interactions across multiple frontostriatal circuits and their implication in the ability to learn abstract task structures that affords generalization and transfer of learned knowledge.
The neural circuit model of the basal ganglia was expanded to explore the role of the subthalamic nucleus (STN), a key target of deep brain stimulation (DBS) surgery in Parkinson’s disease. This model led to a key prediction that the STN was involved in detecting conflict among competing actions represented in cortex, and then exerting an inhibitory effect on behavior, delaying response selection. The model further showed that this amounted to an increase in the “decision threshold” as estimated with more abstract and established mathematical models of decision making. We and others have provided several key tests of this framework, showing that STN activity correlates with adjustments in decision thresholds as a functyon of conflict, and STN disruption leads to reduced decision thresholds, providing a mechanism for the impulsivity observed in patients on DBS (one that is distinct from that induced by dopaminergic medications). We have provided further evidence of this mechanism with neuroimaging and EEG, and developed expanded models of its role – in concert with the prefrontal cortex – in other forms of response inhibition.
A neural network model of the frontal cortex and basal ganglia in executive and inhibitory control. This model captures patterns of electrophysiology observed in multiple brain areas/ cell types as animals and humans are engaged in cognitive control tasks, and provides an account of their roles in behavioral adaptations (error rates, response time distributions). From Wiecki & Frank (2013)
Sequential sample models can be used to quantify dynamics of decision making and cognitive control which can capture the functional characteristics of more detailed neural network models but can also be used to quantitatively fit behavioral data and brain/behavioral relationships.