Fig. 1. Experimental procedure.

a Timeline of a choice trial during the multi-dimensional probabilistic learning task. In each choice trial, the participants chose between two stimuli (colored patterned shapes) and were provided with reward feedback (reward or no reward) for both the chosen and unchosen stimuli. The inset at the top shows the set of all visual features used in the experiment (S: shape; P: pattern; C: color). The inset at the bottom shows the screen during a sample estimation trial. In each estimation trial, the participants provided their estimate about the probability of reward for the presented stimulus by pressing 1 of 10 keys (A, S, D, F, G, H, J, K, L, and;) on the keyboard. b Example of reward probabilities assigned to 27 possible stimuli. Stimuli were defined by combinations of three features, each with three instances. Reward probabilities were non-generalizable, such that reward probabilities assigned to all stimuli could not be determined by combining the reward probabilities associated with their features or conjunctions of their features. Numbers in parentheses demonstrate the actual probability values used in the experiment due to limited number of trials. For the example schedule, the shape was on average informative about reward (average probability of reward on three shapes were equal to 0.3, 0.5, and 0.7). Although pattern and color alone were not informative, the conjunction of these two non-informative features was on average informative about the reward. Each participant was randomly assigned to a condition where the informative feature was either pattern or shape.