El efecto de fiabilidad predictiva en aprendizaje causal humano no se ve afectado por el sobreentrenamiento

Abstract

Previous studies in human causal learning have demonstrated that the predictive history of the stimuli influences their associability (i.e., readiness with which they enter into associations subsequently). Participants receive initial training in which they can learn that half of the cues are accurate predictors of their outcomes, and the other half are poorer predictors (Stage 1). Subsequently, all cues are equally predictive of a new outcome (Stage 2). On test, participants rate the likelihood that the cues would produce Stage 2 outcomes. It has been consistently found that participants rate the cues that were accurate predictors in Stage 1 higher than poorer predictors. A new experiment is reported demonstrating that a reduction of the length of Stage 1 training does not affect the magnitude of this predictive accuracy effect. This finding is inconsistent with the notion that the effect is a consequence of changes in the associability of the cues on Stage 1. An alternative interpretation of the effect in terms of learning and generalization of performance rules of is considered.