Inference more often than false memory guides novel decision making

Memory plays a crucial role in many decisions, allowing individuals to recombine elements of past episodes stored in memory to guide choice (Biderman et al., 2020; Shohamy & Daw, 2015). However, memory is fallible and can get distorted. This begs the question: how do these distortions affect decision making? Here, we explore two non-exclusive potential mechanisms. False memories formed prior to a decision may guide choice (Bui et al., 2013; Storbeck & Clore, 2005; Wang et al., 2024) and individuals may make inferences based on structured knowledge stored in memory at the time of choice (Carpenter & Schacter 2018; Schacter et al. 2011). To better understand whether these mechanisms contribute equally to behavior, we employed a modified version of the classic Deese-Roediger-McDermott (DRM) paradigm (Deese, 1959; Roediger & McDermott, 1995), where we assigned different levels of reward to different DRM lists and different specific values with each word within a list at encoding. We then tested participants’ word recall, their memory for associated rewards, and/or their value-based choices for studied and unstudied words (including critical lures that were related to studied lists). We found that memory for critical lures was often successfully monitored and participants most often declined to provide a specific value they thought was associated with it—unlike old words, which were much more likely to be assigned a value—suggesting that most lures were not embedded within true false memories. Thus, the observed choice bias in favor of high-value critical lures was most often the product of inference at the time of choice rather than driven by false memory of a rewarding experience. We propose that detail-rich false memories for rewarding experiences are seldom formed, but that inferences about potential outcomes based on structured knowledge stored in memory are far more ubiquitous in guiding adaptive choice.