respectively. A significant interaction between session and age group was obtained, F(3,66) = 3.15, p < .05, indicating that as the time under automated control increased, the behavior of the young and old age groups grew increasingly different. From looking at a graph of the means in Figure 1, we can see that the elderly age group shows an immediate drop in detection rate of automation failures from training to session 1. The elderly group then exhibits another drop in the detection rate in session 2 and appears to level off in session 3. This drop is indicative of the classic vigilance decrement exhibited by participants in previous research. The young age group exhibits a decrease in detection rate from training to session 1, a slight rise in detection rate in session 2 and another rise in detection rate in session 3.
Even though the performance for the young and old age groups differed significantly, no significant differences in subjective workload were obtained in this experiment (Figure 2). Subjective workload for both groups across sessions remained relatively unchanged.
The first hypothesis, that older individuals would perform poorer than younger individuals, was supported. The detection rate of both age groups grew increasingly different over time. Detection rate of automation failures was higher for the younger group than for the older group. Detection rate increased significantly for the younger group as a function of time on task, whereas the older group showed a decline over time. Performance cost of automation-induced complacency is more pronounced in the older group than the younger. The second hypothesis that the subjective workload experienced by the older group would be significantly greater than the subjective workload experienced by the younger group was not supported. Subjective workload does not vary as a function of age. Both the older and younger group experienced comparable high levels of subjective workload. Performance under automation control was significantly lower than under manual control. This confirms previous work on automation-induced monitoring inefficiency ( Parasuraman, Molloy, & Singh, 1993). Even though the detection rate of the younger group was significantly higher than for the older group, monitoring of automated systems remains poor regardless of age group. The older group exhibited the expected performance under high workload conditions but the younger group began to improve as a function of time on task indicating that more cognitive resources were available in the younger group or the younger group was able to allocate cognitive resources more efficiently.
One proposed method of maintaining high levels of monitoring performance lies in proper implementation of adaptive task allocation or adaptive automation. Adaptive automation refers to a system capable of dynamic, workload-triggered reallocations of task responsibility between human and machine ( Hilburn, Jorna, Byrne, & Parasuraman, 1997). The traditional approach to allocation of functions in a human-machine system was originally determined by a dichotomous relationship, which gave full control of a task either to the human, or to the machine. In today's technological society it is debatable whether some of the functions that need to be performed are better executed by a human or by a machine ( Fitts, 1951). According to proponents of adaptive systems, the benefits of automation can be maximized and the costs minimized if tasks are allocated to automated subsystems or to the human operator in an adaptive, flexible manner rather than in an all-or-non fashion ( Rouse, 1988). An ideal situation would be one where the operator could switch the control of a task from manual to automated when workload conditions are high. Once the operator's workload was reduced, the operator could then continue to perform the task in manual