Figure 1.

Task structure

(A) Participants were introduced to four people, who were each associated with one animal and one vegetable. Each unique vegetable and animal were associated with two people. Animals and vegetables presented as outcomes were probabilistically associated with a coin worth £1.

(B) On each trial participants chose, within 2 s, one of two randomly offered people who shared one outcome (animal or vegetable) in common, and subsequently obtained as outcomes this person’s favorite animal and vegetable in random succession. At this point, participants learned for each outcome whether it provided a reward (in the current example trial the monkey yielded a reward, indicated by a coin, but not the garlic, indicated by “0”). Outcomes were presented in one of three formats, randomly interleaved with equal probability. The standard presentation format, wherein the identities of both outcomes were exposed, is illustrated in (B).

(C) In the prospective inference format, for the trial illustrated in (B), participants saw a black curtain instead of the second outcome (garlic) but its reward or lack thereof was presented. Importantly, after seeing the first outcome (monkey), participants could predict, based on knowledge of the transition structure (A), the identity of the second outcome. Similarly, in the retrospective inference format, for the trial illustrated in (B), participants saw a curtain instead of the first outcome (monkey). Here, participants could only infer the identity of this first outcome after they saw the second outcome. Note, however, that by that time the reward associated with the hidden outcome was no longer perceptually available. Dashed black arrows indicate the display change necessary to cast the standard trial (B) into an inference format.