Abstract
Imagine you are shown 8 colors, and then probed on your memory. You perform poorly and subjectively feel like you are “guessing”. This intuition is common and is part of the appeal of “slot” models, since it is consistent with discrete limits on how many items are represented. Is such a subjective feeling of guessing incompatible with signal detection and other resource-like accounts that propose you have information about all items? Here we show that *subjective* guessing is naturally and clearly predicted by signal detection accounts. However, subjective guessing is not the same as *objective* guessing (d’=0). We provide strong evidence against objective guessing even at the highest set sizes, including (1) a meta-analysis of previous data used to argue for slots, as well as (2) new continuous report data (N=30) with subjective confidence reports. We also show that signal detection accounts of these data (Schurgin, Wixted, & Brady, 2018) perfectly predict the phenomena of subjective guessing, since confidence in signal detection is derived from a single sample of underlying memory strength, which varies item-to-item. What does it mean to have some information about all items in working memory? Typical models and intuitions rely on the idea that each item has a single representation at any given moment (I think the item is green, but it was yellow). Signal detection models instead conceptualize memory in terms consistent with population codes (eg., Bays 2014) or a probability distribution over colors. That is, having some information about an item simply means having (probabilistically) ever-so-slightly higher activity in sensory channels related to colors that were previously seen, compared to colors that were not previously seen at a given location. Altogether, our results demonstrate how subjective guessing is predicted by signal detection models and can falsely lead to intuitions that favor slot models.
Acknowledgement: NSF CAREER BCS-1653457 to TFB