No abstract

The ability to learn not only from experienced but also from merely fictive outcomes without direct rewarding or punishing consequences should improve learning and resulting value-guided choice. Using an instrumental learning task in combination with multiple single-trial regression of predictions derived from a computational reinforcement-learning model on human EEG, we found an early temporospatial double dissociation in the processing of fictive and real feedback. Thereafter, real and fictive feedback processing converged at a common final path, reflected in parietal EEG activity that was predictive of future choices. In the choice phase, similar parietal EEG activity related to certainty of the impending response was predictive for the decision on the next trial as well. These parietal EEG effects may reflect a common adaptive cortical mechanism of updating or strengthening of stimulus values by integrating outcomes, learning rate, and certainty, which is active during both decision making and evaluation. Neuronal processing of real (rewarding, punishing) and fictive action outcomes (which would have happened had one acted differently) differs for 400 ms and then converges on a common adaptive mechanism driving future decision making and learning.

Following recent advances in neuromodulation therapy for mental disorders, we treated one patient with severe alcohol addiction with deep brain stimulation (DBS) of the nucleus accumbens (NAc). Before and one year following the surgery, we assessed the effects of DBS within the NAc on the addiction as well as on psychometric scores and electrophysiological measures of cognitive control. In our patient, DBS achieved normalization of addictive behavior and craving. An electrophysiological marker of error processing (the error-related negativity) linked to anterior mid-cingulate cortex (aMCC) functioning was altered through DBS, an effect that could be reversed by periods without stimulation. Thus, this case supports the hypothesis that DBS of the NAc could have a positive effect on addiction trough a normalization of craving associated with aMCC dysfunction.

Adapting to errors quickly is essential for survival. Reaction slowing after errors is commonly observed but whether this slowing is adaptive or maladaptive is unclear. Here, we analyse a large dataset from a flanker task using two complementary approaches: a multistage drift-diffusion model, and the lateralisation of EEG beta power as a time-resolved index of choice formation. Fitted model parameters and their independently measured neuronal proxies in beta power convergently show a complex interplay of multiple mechanisms initiated after mistakes. Suppression of distracting evidence, response threshold increase, and reduction of evidence accumulation cause slow and accurate post-error responses. This data provides evidence for both adaptive control and maladaptive orienting after errors yielding an adaptive net effect – a decreased likelihood to repeat mistakes. Generally, lateralised beta power provides a non-invasive readout of action selection for the study of speeded cognitive control processes.