Increasing predictive estimations without further learning the peak-shift effect

The peak of learned responding normally occurs at the learning stimulus itself, but can shift to a different stimulus after discriminative learning. This provides important information about the nature of the generalization mechanism, and reveals alternative pathways through which learned responses can increase. Over two experiments, we established the peak-shift effect in a human predictive learning paradigm. Participants were asked to predict the occurrence of a neutral outcome (drawing of a lightning bolt) based on preceding geometrical figures (rings of different sizes). During learning, the middle-sized ring was sometimes followed by the outcome, whereas a larger ring was never followed by the outcome. At test, we presented larger and smaller rings (Experiment 1), or only a slightly smaller ring (Experiment 2). We consistently observed highest prediction of the outcome to the slightly smaller ring. Predictive estimations in humans can reach their height to stimuli that have never actually participated in the learning experiences. We argue that the results are most in line with an associative learning account, rather than an adaptation-level or a rule-learning account. © 2013 Hogrefe Publishing.