Proceedings of the Annual Meeting of the Cognitive Science Society

Humans readily perceive patterns even in situations with only random noise, but how do they respond when the genuine rule deliberately discourages this tendency? We investigated superstitious ordering under a consistent preference-discouraging (PD) schedule, in which reward probability increases for previously avoided options and decreases for previously chosen options. This contradicts typical reinforcement principles, in which some options are always preferred over others. We also examined the role of visual features, with one group using geometric shape combinations as stimuli and another group using natural images. We found that participants in both groups developed significant subjective linear orderings of stimuli, with a marginally stronger effect in the Geometrical Image condition. Comparisons with the benchmark agent, optimized Q-learning agent, and random agent showed that none of the models could explain participants' response patterns, indicating a unique and stable superstitious ordering effect in human learning.