Steven A. Murray
SPA WAR Systems Center
San Diego, CA
Barrett S. Caldwell
University of Wisconsin
Increased reliance on automation in the design of human-machine systems is based largely on the assumption that reducing task demands will reduce the probability of human overload and lead to more reliable system performance. Automated systems, however, shift the operator's job from continuous control to intermittent or supervisory control, inducing boredom and monotony; personal involvement with the state of the system is also reduced. The supervisory control environment can therefore diminish the operator's ability to respond to abnormal conditions following extended periods of work underload, creating a new path for system failure (e.g., Ryan, Hill, Overline, & Kaplan, 1994).
Krulewitz, Warm, and Wohl ( 1975), using a variable-demand vigilance task, found that subjects transitioning from a slow to a fast event rate performed more poorly in the immediate post-transition period than control subjects who had performed the task at the fast rate throughout the test session. This result indicated that a shift between two workload levels may impose performance costs over and above those of either workload level alone; i.e., resources are required for the adapting process itself.
Our research focused on the performance of workstation operators under conditions of protracted work underload, so the general issue of supervisory control and the specific results of the Krulewitz, et al. ( 1975) study were of interest to us. The literature of vigilance performance (e.g., Davies & Parasuraman, 1982) has shown that human capabilities vary in both tonic and phasic fashion, over a range of time scales, and that these fluctuations can become more prominent in settings with little external stimulation. It was reasonable to expect that such operator fluctuations would influence the degree of the efficiency of the adapting process found by Krulewitz, et al., and others. We were specifically interested in evaluating performance changes to high workload events as a function of the operator alertness states that preceded those events. If the availability of cognitive resources was a function of operator alertness during work underload, then task performance should be directly related to measures of operator alertness state during these underload periods. In particular, higher rates of performance errors would be related to lower levels of pre-task alertness, as operators failed to attend to task components that they could normally handle.
An air defense scenario was chosen for the experiment, typical of the kind performed in the Navy and elsewhere, and included both a primary task and a secondary task. The primary task was to process aircraft inbound to two ship icons, shown at fixed positions on a conventional color monitor. Aircraft symbols originated from random points around the borders of the screen and proceeded inward toward the ships. The subject's job was to classify each aircraft symbol as a friend or an enemy, to destroy enemy aircraft before they advanced too close to the ships, and to avoid shooting down any friendly aircraft. Subjects classified aircraft by selecting a symbol with a trackball, matching its displayed name to a list of friends and enemies, and designating it with a pair of pushbuttons. The secondary task required the subject