The effects of learned predictiveness and uncertainty on associability

Abstract

The Mackintosh (1975) and Pearce-Hall (1980) models of learning propose that prior experience with a cue affects the attention paid to it, but they differ in their predictions for how attention will change. In human learning tasks, it is commonly found that good predictors of outcomes receive more attention than poor predictors (learned predictiveness). In this thesis, however, the opposite result was found: a cue which leads to an uncertain outcome received more attention (learned uncertainty). Surprisingly, this effect relied on the level of difficulty of training procedure. Attention was allocated to uncertain cues when the training procedure used few uncertain compounds; while attention paid to predictive cues was higher than uncertain cues when the training procedure used more uncertain compounds. These results may reflect that participants engage in different processes of automatic attention and controlled attention based on the difficulty of the task. Moreover, the pattern of these results was also found in the comparison between cues previously used in a biconditional discrimination and predictive cues. When the difficulty of the training procedure was relatively low, biconditional cues received more attention than predictive cues, and the opposite was found when the difficulty of the training procedure was relatively high. Biconditional cues are similar to uncertain cues in that individual cues are uninformative, but they differ from uncertain cues in that they are informative when presented in compound. Therefore, the fact that biconditional cues underwent changes in attention that were similar to uncertain cues suggest that the changes in attention depend on the individual prediction error term rather the than summed prediction error term.