Real and Fictive Outcomes Are Processed Differently but Converge on a Common Adaptive Mechanism

SummaryThe 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