CENTER FOR RESEARCH IN
COGNITION & NEUROSCIENCES/ConnexConnex

Tasks that require cognitive control reliably engage a cluster of regions including anterior cingulate cortex (ACC), dorsolateral prefrontal cortex (DLPFC), and anterior insula (AIC). Despite the ubiquity of studies reporting control-relevant activity in these regions, research has thus far failed to converge on a consensus regarding the role played by each individual region, much less how they collaborate to identify the need for and deploy control. In this talk, I will describe my ongoing work in developing and testing computational neural models that aim to provide a comprehensive framework for explaining the function of control-related regions. Recent models of ACC and DLPFC suggest how these two regions interact to support control using a common neural currency of error processing: ACC predicts future events and reports deviations between observations and expectations, while DLPFC learns representations of error signals generated by ACC in order to improve subsequent predictions. Together, the Predicted Response-Outcome (PRO) and Hierarchical Error Representation (HER) models account for a vast array of effects observed in ACC and DLPFC across a variety of experimental contexts and at multiple levels of description (single-unit, fMRI/EEG, and behavioral). Tests of the PRO and HER models using fMRI support an error-based account of ACC function over value-based frameworks, and are consistent with distributed representations of rules and task sets in DLPFC. Finally, I will discuss preliminary results suggesting how the PRO/HER framework may be extended to incorporate additional control regions involved in representing uncertainty, particularly AIC.