What is a choice in reinforcement learning?

AbstractIn reinforcement learning (RL) experiments, participants learn to associate stimuli with rewarding responses. RL models capture such learning by estimating stimulus-response values. But what is a response? RL algorithms can model any response type, whether it is a basic motor action (e.g. pressing a key), or a more abstract, non-motor choice (e.g. selecting pizza at the restaurant). Are these different responses learned the same way? In this study, we examine differences between learning a rewarding association between (1) a stimulus and a motor action and (2) two stimuli. We show that learning differs between these two conditions, contrary to the common implicit assumption that response type does not matter. Specifically, participants were slower and less accurate in learning to select a rewarding stimulus. Using computational modeling, we show that the values of motor actions interfered with the values of stimulus responses, resulting in more incorrect choices in the latter condition.


Return to previous page