Learned Error Matching in the Anterior Cingulate Cortex

Joshua W. Brown, Todd S. Braver
Washington University
Department of Psychology
Campus Box 1125
St. Louis, MO, 63130-4899, USA

 

The anterior cingulate cortex (ACC) and related medial wall play a critical role in recruiting cognitive control. However, it is unclear how ACC acquires selective error or conflict effects, particularly when these depend on task context. We propose a new error-matching hypothesis of ACC function: the ACC response to a given task condition is proportional to the perceived likelihood of an error in that condition. Since conflict generally leads to greater error likelihood, conflict-detection is a special case predicted by the error-matching hypothesis. We used computational models to simulate both the error-matching hypothesis and the conflict hypothesis in the context of a modified stop-signal task. Having fit both models to corresponding human behavioral data, we show that the models make competing predictions regarding ACC effects. In particular, only the error-matching model predicts greater ACC response for trials in which errors are more likely, even despite the absence of response conflict or actual errors in a given trial. We tested the competing simulation predictions with an fMRI study of the same task. The fMRI results reveal ACC effects of error-likelihood, even in the absence of conflict or errors. Furthermore, we found evidence that dopaminergic reinforcement signals train performance monitoring and cognitive control mechanisms in the ACC and related medial wall areas. Overall, these results support a new and more general error-matching theory of ACC function based on reinforcement learning, of which conflict and error detection are special cases, and they suggest a clearer link between reinforcement learning and cognitive control.