Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Humans automatically assign credit to outcome-irrelevant features

Ido Ben-Artzi¹, Franz Wurm², Nitzan Shahar¹; ¹Tel Aviv University, ²Leiden University

Presenter: Ido Ben-Artzi

Adaptive behavior depends on correctly assigning credit to the features of an action that caused a reward while ignoring those that did not. Nevertheless, recent work shows that humans also assign credit to features of their actions that are explicitly known to be outcome-irrelevant. Here, we tested the neural basis of this phenomenon by combining EEG and computational modeling of behavior. Forty participants performed a reinforcement learning task in which rewards depended only on card identity, whereas cards’ spatial location was random and explicitly instructed as outcome-irrelevant. Replicating prior findings, rewards increased the likelihood of repeating the previously chosen location even when two different cards than before were offered. Using a computational model, we estimated feature-specific prediction errors for both outcome-relevant cards and outcome-irrelevant locations. Trial-level regression and cluster-based permutation analyses revealed that both prediction errors, as well as their interaction, were encoded in outcome-locked EEG activity. These results suggest that higher-level knowledge of the environment constrains learning less than previously assumed.

Topic Area: Decision-Making, Cognitive Control & Event Cognition