Prediction error and overt attention to relevant and irrelevant cues: Evidence for an interaction of two associability mechanisms
Humans and other animals use cues in the environment to make predictions about important outcomes, thus preparing themselves to respond to those events. Prediction error refers to the extent to which an outcome is surprising in the presence of one or more cues. Within the research area of associativ...
|Online Access:||PDF Full Text|
No Tags, Be the first to tag this record!
|Summary:||Humans and other animals use cues in the environment to make predictions about important outcomes, thus preparing themselves to respond to those events. Prediction error refers to the extent to which an outcome is surprising in the presence of one or more cues. Within the research area of associative
learning, some theories suggest that prediction error changes the amount of attention paid to cues. It was initially proposed that the attentional changes were driven by either relative or overall prediction error. In the first case, attention increases for cues generating less prediction error than other concurrent cues, otherwise attention decreases (Mackintosh, 1975). In the second case, the amount of attention paid to each cue is directly related to overall prediction error, i.e. how surprising the outcome is considering all the present cues (Pearce & Hall, 1980). Evidence for the role of relative prediction error comes from studies with pairs of cues including a component relevant to outcome prediction, together with an irrelevant component. Evidence for a role of overall prediction error comes from studies in which cues generating different amounts of prediction error are trained separately. Given that considering both relative and overall prediction error may account for a wider range of attentional changes, the two mechanisms were incorporated into hybrid models (e.g., Le Pelley, 2004). However, evidence for those models in humans is still scarce. The aim of the present thesis was to study the effect of a sudden rise in overall prediction error on overt attention to cues that were either relevant or irrelevant to outcome prediction, i.e. differing in terms of relative prediction error. Rather than considering sustained levels of prediction error, we focused primarily on sudden changes, because they are involved in important behavioral phenomena, such as the return of pathological anxiety. Each of the two studies included in the thesis started with a discrimination training, in which participants had to predict the occurrence of two possible outcomes. Participants’ eye gaze showed that the relevant cues received more attention than the irrelevant cues. In a second stage, contingency reversal (Study I) or partial reinforcement (Study II) led to a rise in prediction error, as indicated by a drop in the accuracy of outcome predictions. The attentional preference for the relevant cues was temporarily weakened by contingency reversal, and it was completely lost following the introduction of partial reinforcement. In addition, both manipulations increased the amount of attention paid to both types of cues. The data were consistent with a combined effect of relative and overall prediction error, thus providing evidence for the hybrid models. In addition, the results have implications for understanding changes in attention to contextual cues.|
|Physical Description:||97 Pages|