Fig. 1.

Experimental design. We conducted three independent experiments, in which a conditioning block (A–C) was followed by a decision making block (D and E). The conditioning block of the experiments consisted of (A) Pavlovian (classical) threat conditioning (experiment 1), (B) observational threat learning (experiment 2), or (C) instructed threat learning (experiment 3). In the conditioning block, one stimulus (CS+), but not another (CS−), was paired with shocks for the participant (experiment 1), for another person (experiment 2), or verbally associated with the possibility of shock (experiment 3) (Methods). Next, participants made 70 decisions between the same two cues (CS+, CS−), which were both probabilistically punished with electric shocks to the participants (see SI Appendix, Fig. S1 for a trial timeline). In each experiment, the participants were divided into two subgroups. The shock probabilities differed between the (D) No Change groups, and (E) the Change groups. In the No Change groups, the CS+ from the conditioning block had a higher probability of being followed by shocks in the Transfer phase of the decision making block, while in the Change groups, the CS+ had a lower probability of being followed by shocks. After 35 trials, the outcome contingences reversed (Reversal phase). CS, conditioned stimulus.