Search Papers | Poster Sessions | All Posters
Poster A158 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Perseverative Behavioral Sequences Aid Long-Term Credit Assignment
Sienna Bruinsma1 (), Frederike Petzschner1,2, Matthew Nassar1; 1Carney Institute for Brain Science, Brown University, 2Department of Psychiatry & Human Behavior, Alpert Medical School of Brown University
Learning from the past requires assigning credit to the consequences of our actions. Here, we explored human credit assignment strategies by asking people to select activities with short- and long-term pain-related consequences for an avatar and predict the avatar’s subsequent pain level. Human behavioral results suggest that, while participants can learn short-term consequences, their learning of long-term consequences depends critically on how they sequence activities in time. More specifically, increased repetition in activity selection (i.e., perseveration) is related to a learned preference for activities that reduce long-term, but not short-term, pain. Additionally, in comparing several computational models, we found that standard model-free algorithms (i.e., temporal difference learning) best explained the behavior of participants who did not perseverate, whereas Bayesian inference models that take into account the causal structure of the environment better explained those that did. Our results demonstrate that credit assignment critically depends on the order in which actions are selected, with repetitions aiding the learning of long-term consequences. This raises the possibility of perseveration as a useful action policy to improve long-term credit assignment.
Keywords: credit assignment perseveration temporal difference learning Bayesian causal inference