PROJECT SUMMARY Imagine having to choose between an apple or an orange for a snack. To decide between these famously incomparable options, you may think back to previous experiences you have had with either fruit and choose the one you enjoyed the most. To make a choice in this manner, it is crucial that you remember your previous snacks with a relatively high level of detail, specifically when it comes to the identity of the fruit. Conversely, imagine a decision-maker who is asked to choose between two novel options: a new brand of chocolate, and a fruit they have never tasted before. In this case, choosing the right snack relies on abstracting over the details of past experiences to infer the high-level value of chocolate versus fruit – the specifics of each experience are no longer relevant. Thus, less precise memory can facilitate the generalization of value to novel exemplars of a particular kind, while hindering people’s ability to choose between similar options. Here, we propose to test this hypothesis by leveraging the decreases in memory precision that occur naturally in healthy aging. We will collect fMRI data while participants take part in two variants of a decision task that relies on estimating value from past experience in service of choice. In Aim 1, participants will choose between items that are semantically and perceptually similar, meaning that precise memory is necessary for accurate choice. We predict that older adults will perform poorly at this task because of decreases in the precision with which they can remember stimulus identities. Furthermore, we predict that decreased pattern separation in the posterior hippocampus – a phenomenon by which similar items are kept separate in memory via enhanced neural differentiation – will be at the root of this effect. In Aim 2, participants will be asked to make choices between novel exemplars of specific categories whose value they can learn over time. We hypothesize that older adults, because of this same decrease in memory precision, will actually generalize category value faster than their younger counterparts, who may be more biased by noise from individual exemplars. Neurally, we expect to find greater pattern completion (the opposite of pattern separation) in the hippocampus of older adults, as well as increased connectivity between the hippocampus and other brain regions known to track category information (ex: medial prefrontal cortex). We will also characterize these effects computationally by comparing the performance of an “object only” reinforcement learning (RL) model in which each exemplar is assigned a separate value to a model that understands categories by separating them into discrete states towards which value can be assigned.