A Processing Ghost in a Tank Machine
College of Liberal Arts and Sciences
Social and Behavioral Sciences
Two major contrasting models of visual working memory (WM) have been prevalent in recent cognitive literature. The discrete-slot model proposes that WM operates on the all-or-none principle: holding only high-resolution item representations stored in a limited number of memory slots. Items exceeding a storage capacity limit have zero-resolution representations. According to the variable-resources model, WM operates on the all-get-some principle: a pool of limited resources is dynamically allocated across a set of memorized items representations. WM can potentially hold an unlimited number of items by lowering their resolution. The most recent advancements in the theory have amassed evidence supporting the all-or-none discrete-slot WM model, while providing little evidence to support the variable-resource model. These findings could be use to imply that underlying neural representations need to be all-or-none. However, we argue that the aforementioned research advances have been downplaying experimental approaches that directly manipulate the allocation of resources across item representations held by WM. The results of the present study indicate that when given certain instructions, subjects in the experiment adaptively allocated a limited amount of resources and shared them across memorized item representations. The study furthermore demonstrates that the cognitive system operates under the principle of resource conservation: Allocating more resources to some item representations led to allocating fewer resources to the other item representations. In turn, each memorized item representation could show a gradual change from low to high-resolution states. The computational modeling results, conducted jointly on the reaction time and accuracy data, support the idea that a simple linear function could be used as a proxy to the underlying resource allocation across memorized item representations, very much as the water level in a tilted tank behaves. Overall, the findings can be well explained by the variable-resource model of WM, and are inconsistent with the discrete-slot and the so-called dual approaches, postulating that WM is partitioned into the two qualitatively different storages: one that allows direct and fast access to maintained item representations, and the other which allows retrieval of stored items by a slower search process. The results of the adjoined computational modeling experiment suggest that that the cognitive system can operate with low-resolution (noisy) WM representations of stored items, thus implying that neural representations do not need to be all-or-none.
4th Annual Midwest Cognitive Science Conference
Fific, Mario, "A Processing Ghost in a Tank Machine" (2014). Faculty Scholarly Dissemination Grants. 919.