Proceedings of the Annual Meeting of the Cognitive Science Society

n reinforcement learning (RL) experiments, participantslearn to associate stimuli with rewarding responses. RLmodels capture such learning by estimating stimulus-responsevalues. But what is a response? RL algorithms can model anyresponse type, whether it is a basic motor action (e.g. pressinga key), or a more abstract, non-motor choice (e.g. selectingpizza at the restaurant). Are these different responses learnedthe same way? In this study, we examine differences betweenlearning a rewarding association between (1) a stimulus and amotor action and (2) two stimuli. We show that learningdiffers between these two conditions, contrary to the commonimplicit assumption that response type does not matter.Specifically, participants were slower and less accurate inlearning to select a rewarding stimulus. Using computationalmodeling, we show that the values of motor actions interferedwith the values of stimulus responses, resulting in moreincorrect choices in the latter condition.