Figure 2.

Probabilistic selection (PS) task. (A) Each trial in the PS task began with a jittered fixation cross followed by a pair of stimuli for 2500 ms. Following a left or right button response, the selected image appeared highlighted on the screen for the duration of the 2500 ms presentation. Choices during training then received probabilistic feedback, whereas those during testing were followed by the fixation cross marking the next trial. (B) In the training phase, participants learned to choose between three discrete pairs of Japanese Hiragana characters with different reward contingencies. Each pair had a better and worse choice, but the relative weights of these values changed. The reward probabilities of each stimulus are shown in parentheses. (C) In the test phase, participants generalized their knowledge of the training pairs to recombined stimulus pairs. There was no feedback in this phase. Learning to choose A over B during training could reflect approach learning, avoidance learning, or both, and so we assessed overall test performance as well as the accuracy of (A) choices and (B) avoidances when these stimuli appeared in novel pairs during testing.