Attention, Resource Models

Evolution of Models

The notion of resources explicates a premise implicit in naive thinking and reflected in natural language: Attention is said to be “paid,” namely, directed and/or expended. Whatever is expended must be taken to be some sort of structure or commodity that is limited in some way, hence internal input. Early theorists of attention, such as Alan Welford and Donald Broadbent, construed input, respectively, as a uni-partite channel that may only be directed as a whole to a task or an object of attention, or as a single channel with limited capacity that is typically capable of handling only a single task.

Daniel Kahneman posited that internal input may be expended in various amounts and may be divided between objects of attention. He argued that the existence of such input is suggested by what he called the intensive aspects of attention, namely, the dependence of performance on vigilance or arousal, which he ascribed to the degree of accessibility or activation of that internal input.

Donald Norman and Daniel Bobrow coined the term resources as a label for the hypothetical internal input assumed to be demanded for a broad range of functions believed to require attention. They suggested that performance of a task or process is resource limited, namely, that it depends on the amount of resources invested in it, up to a point at which performance becomes data limited, namely, insensitive to resources and determined just by data quality. For example, a task of detecting a moderately loud tone may demand a small quantity of resources; when that quantity is not fully available, detection may sometimes fail; when it is available, failures must be due to poor sensory data. However, the shape of this performance-resource function is strongly task specific. The way Norman and Bobrow proposed to study that function is via analyzing dual-task performance by means of a curve they termed performance operating characteristics (POCs), plotting performance of one task as a function of performance of the concurrent task.

...