Complex tasks frequently require holding in mind several ideas at once. For example, when performing mental addition we must juggle several numbers in mind while applying operations to update those numbers. It is well-established that there are distinct limits to the amount of information that can be held in mind at any given time and that these limits have a close relationship to our abilities to reason and to comprehend -- in short, to our intelligence. However, the fundamental architecture of the short-term memory system responsible for these capacity limits remains unclear. Recent psychological research suggests that distinctions can be drawn between a single piece of information at the forefront of the mind, other information that is being held in mind, but currently not the focus of attention, and information that is not being held in mind, but available in long-term memory. The goal of this project is to understand the brain mechanisms responsible for representing these different states of information. With support from the National Science Foundation, Dr. John Jonides and colleagues at the University of Michigan will investigate the architecture of short-term memory using functional magnetic resonance imaging. The research will examine neural responses as human volunteers retain, retrieve, and update information in the various putative states of short-term memory. In addition to traditional measures that afford brain localization of function, the research will employ analyses that examine functional networks of brain activation, thus elucidating brain mechanisms by which short-term memory is achieved. The work will also examine neural responses to different types of information, such as verbal, spatial, and pictorial information and examine the brain networks involved in subserving short-term memory of each form.
Since much of human intelligence depends on capacity limits of short-term memory, understanding how the brain implements short-term memory could have substantial impact on cognition as a whole. The understanding furthered by this research may inform methods that aim to increase short-term memory capacity and intelligence. This research project will also provide comprehensive training functional magnetic resonance imaging techniques to undergraduate and graduate students involved in the project. Findings from this study will be disseminated widely through publications and a seminar.
Consider the act of computing 421 x 3 + 67 in your mind. You might begin with "421 x 3" and add "67" afterwards in order to preserve the order of operations. Then, to calculate "421 x 3", you might sequentially compute each sub-part of the operation (i.e. "1 x 3" then "2 x 3" then "4 x 3"). This formulation suggests a trisection of maintenance: items that are passively maintained to be returned to at a later time ("+ 67"), items that are actively maintained to be operated upon ("421 x 3"), and items that are the current focus of operation ("1 x 3"). How do we perform this mental juggling act? A simple model suggests that during processes such as mental arithmetic, all of the information is held in mind in a similar way with different items distinguished by a single parameter such as memory strength. This model predicts that neural computations that access and manipulate different items in working memory should be similar, but graded as a function of memory strength. Quite a different model suggests that there are distinct representational states (e.g. focused, active, passive) that allow us to arrange and distribute mental resources in order to accomplish complex cognition. This model predicts distinct neural signatures associated with accessing information in different states of representation. In a series of studies, we have used functional MRI to study the neural underpinnings of short-term memory and have found dissociable neural signatures associated with the access to focused, active, and passive information supporting the distinct state model. This triple dissociation suggests that working memory is trisected into distinct representational states. A three-state model of memory provides a useful framework to understand the interactions among attention, working memory, and long-term memory. While each of these domains has often been studied in isolation, we have advocated a model that hypothesizes that these domains map onto distinct states that simultaneously contribute to short-term maintenance. The model explains how the contents of working memory can direct externally oriented attention, and how working memories evolve into longer-term traces. Furthermore, the model predicts which operations can be performed in parallel with little interference and which operations conflict with one another. Hence, the model is useful not only to explain memory, but also complex cognitive processes such as dual-tasking.
