Fig. 2

Behavioral results—PE directly determines the degree of learning from feedback and adoption of false memories. a Subsequent accuracy in the recall study. For incorrectly answered Test1 questions (green), the greater the PE elicited by feedback, the more likely it was for that information to be incorporated into memory, as shown by average Test2 accuracy (left). Similarly, for correct Test1 answers (blue), false-feedback information was more likely to supplant an accurate memory recalled with medium confidence (medium PE) relative to one recalled with low confidence (low PE), as shown by the average percentage of false-feedback answers subsequently supplied in Test2 (right). b Subsequent confidence of updated memories in the recall study. PE also determined the confidence expressed in correct Test2 recollections, for questions initially answered erroneously in Test1. Thus, when successfully learning from feedback, incorrect Test1 answers stated with high confidence (high PE feedback) were associated with high-confidence correct answers in Test2 and vice versa (average Test2 confidence for incorrect-to-correct answers on the left). Similarly, for the subset of questions answered correctly and supplanted with false feedback, average Test2 confidence in the false memory was greater when initial confidence in the correct answer was higher (medium vs. low PE, on the right). c Summary of behavioral results in the recognition fMRI study. A strong positive relationship was observed between the degree of overall memory updating and PE arising from feedback, for questions answered incorrectly in Test1. This measure of learning combines both subsequent accuracy and confidence (see Supplementary Figure 3 for separate measures). Error bars represent SEM