Biderman, N., Shohamy, D.
Nat Commun 12, 4648 (2021).
https://doi.org/10.1038/s41467-021-24907-x
Abstract
The goal of deliberation is to separate between options so that we can commit to one and leave the other behind. However, deliberation can, paradoxically, also form an association in memory between the chosen and unchosen options. Here, we consider this possibility and examine its consequences for how outcomes affect not only the value of the options we chose, but also, by association, the value of options we did not choose. In five experiments (total n = 612), including a preregistered experiment (n = 235), we found that the value assigned to unchosen options is inversely related to their chosen counterparts. Moreover, this inverse relationship was associated with participants’ memory of the pairs they chose between. Our findings suggest that deciding between options does not end the competition between them. Deliberation binds choice options together in memory such that the learned value of one can affect the inferred value of the other.
From the Discussion
We found that stronger memory for the deliberated options is related to a stronger discrepancy between the value assigned to the chosen and unchosen options. This result suggests that choosing between options leaves a memory trace. By definition, deliberation is meant to tease apart the value of competing options in the service of making the decision; our findings suggest that deliberation and choice also bind pairs of choice options in memory. Consequently, unchosen options do not vanish from memory after a decision is made, but rather they continue to linger through their link to the chosen options.
We show that participants use the association between choice options to infer the value of unchosen options. This finding complements and extends previous studies reporting transfer of value between associated items in the same direction, which allows agents to generalize reward value across associated exemplars. For example, in the sensory preconditioning task, pairs of neutral items are associated by virtue of appearing in temporal proximity. Subsequently, just one item gains feedback—it is either rewarded or not. When probed to choose between items that did not receive feedback, participants tend to select those previously paired with rewarded items. In contrast, our participants tended to avoid the items whose counterpart was previously rewarded. Put in learning terms, when the chosen option proved to be successful, participants’ choices in our task reflected avoidance of, rather than approach to, the unchosen option. One important difference between our task and the sensory preconditioning task is the manner in which the association is formed. In both tasks a pair of items appears in close temporal proximity, yet in our task participants are also asked to decide between these items and the act of deliberation seems to result in an inverse association between the deliberated options.