Proceedings of the Annual Meeting of the Cognitive Science Society

To act effectively in a complicated, uncertain world, peopleoften rely on task-sets (TSs) that define action policies over arange of stimuli. Effectively selecting amongst TSs requiresassessing their individual utility given the current world state.However, the world state is, in general, latent, stochastic, andtime-varying, making TS selection a difficult inference for theagent. An open question is how observable environmentalfactors influence an actor's assessment of the world state andthus the selection of TSs. In this work, we designed a noveltask in which probabilistic cues predict one of two TSs on atrial-by-trial basis. With this task, we investigate how peopleintegrate multiple sources of probabilistic information in theservice of TS selection. We show that when action feedback isunavailable, TS selection can be modeled as “biased Bayesianinference”, such that individuals participants differentiallyweight immediate cues over TS priors when inferring thelatent world state. Additionally, using the model’s trial-by-trial posteriors over TSs, we calculate a measure of decisionconfidence and show that it inversely relates to reactiontimes. This work supports the hierarchical organization ofdecision-making by demonstrating that probabilistic evidencecan be integrated in the service of higher-order decisions overTSs, subsequently simplifying lower-order action selection.