Figure 2: Ticket bandit task.

(a) The ticket-bandit task. Each slot machine (‘bandit’) delivered tickets—trial-unique photographs—associated with a dollar value—either −$5 or $5. (b) Payoff probabilities. The probability of each bandit paying out a winning ticket varied slowly over the course of the experiment. Participants were told that their total payout would be contingent both on the number of winning tickets they accrued and their ability to correctly respond on a post-task memory test asking them to recall the reward value and slot machine associated with each ticket. (c) Memory probes. Participants encountered 32 recognition memory probes. On 26 of these probe trials, participants were shown objects that were either received on a previous choice trial (‘valid’), whereas on others they were shown new objects that were not part of any previous trial (‘invalid’). Participants were asked only to perform a simple old/new recognition judgement—to press ‘yes’ if they had seen the image previously in this task and ‘no’ if they had not. After each recognition probe, the sequence of slot machine choices continued as before.