Biased Guessing in a Complete-Identification Visual-Working-Memory Task: Further Evidence for Mixed-State Models

Research is reported that provides evidence for a significant role of mixed states and guessing processes in tasks of visual working memory (VWM). Subjects engaged in a complete-identification VWM task. The stimulus set consisted of 16 colors roughly equally spaced around a color circle. On each trial, a memory-set drawn from the colors was briefly presented, followed by a location probe. Subjects attempted to reproduce the color of the probed item by clicking on the appropriate response button of a discrete color wheel. The key manipulation was to vary payoffs for alternative correct responses across trials. Analysis of the resulting matrices of individual-subject identification-confusion data provided evidence for a systematic guessing process: On trials in which subjects had no memory for the probed stimulus, they guessed with high probability using the high-payoff response. Formal modeling corroborated this interpretation. Mixed-state models that assumed that performance involved a combination of memory-based responding and biased guessing yielded accurate and easy-to-interpret accounts of the identification data; by comparison, variable-resources (VR) models without a guessing state struggled to account for the data, including versions with bias parameters for the high-payoff response. The authors argue that the work adds to recent converging sources of evidence that point to a significant role of discrete, mixed states in VWM. The authors also suggest directions for development of extended VR models with sophisticated knowledge-rich decision rules for the complete-identification task. Public Significance Statement Visual working memory serves as a foundation for numerous cognitive processes and tasks, including the ability to search for visual targets, to comprehend visual displays, and to detect changes in visual scenes. These issues are of great significance to the Air Force Office of Scientific Research (the source of funding for the present work). For example, limits on visual working memory are of crucial importance in areas such as monitoring of unmanned aerial vehicles and the piloting of aircraft. In this work, we find evidence that the limits on visual working memory arise from limits in the number of objects that a human can store in memory. The evidence suggests that zero information for the unstored items is retained. Such a result has profound implications for what types of visual-display systems would lead to optimal forms of human perf