Todd M. Eischeid Mark W. Scerbo Frederick G. Freeman Old Dominion University
Automated systems that can shoulder changing responsibilities are generally considered adaptive. ( Hancock & Chignell, 1987; Morrison, Gluckman & Deaton, 1991; Rouse, 1976). Adaptive automation allows the level or mode of automation or the number of systems currently automated to be modified in real time in response to situational changes ( Scerbo, 1996). Parasuraman, Bahri, Deaton, Morrison, and Barnes ( 1992) state that adaptive automation produces an ideal coupling of the level of automation to the level of operator workload. Moreover, adaptive automation represents an attempt to maintain an optimal level of operator engagement.
One adaptive system, a biocybernetic closed-loop system developed at NASA ( Pope, Bogart, & Bartolome, 1995; Pope, Comstock, Bartolome, Bogart, & Burdette, 1994), uses biofeedback to avoid hazardous states of operator awareness. The closed-loop system used by Pope et al. ( 1994; 1995) is truly adaptive and changes tasks in real time. The criterion for the changes in automation is an index of operator engagement [20 beta/(alpha+theta)], derived from relative powers of various bandwidths of EEG activity. The present study also used this index to determine operator engagement.
The idea of a computer as a teammate has been explored by Scerbo, Ceplenski, Krahl, & Eischeid ( 1996), who compared the performance of human-human and human-computer teams. In that study, a pursuit tracking task was partitioned into horizontal and vertical axes, and control of each axis was allocated to a member of the team. In addition, the skill level of the computer was manipulated. Participants paired with an expert computer outperformed all other participants, showing that the skill level of one's teammate affected performance. The human-human team exhibited the most improvement over time, but never did reach the high level of performance of participants paired with an expert computer. Scerbo et al. ( 1996) concluded that partitioning may be a desirable strategy for adaptive automation.
It is important to note that the tracking task used by Scerbo et al. ( 1996) was not adaptive in nature. The present study addressed this gap by utilizing the biocybernetic closed-loop system.
Prior to the present study, task partitioning had not been tested in the closed-loop environment. The tracking task of the MAT was partitioned into horizontal and vertical axes similar to Scerbo et al. ( 1996). Control of one axis was allocated to or taken away from the operator (by the biocybernetic system) depending upon level of engagement and feedback conditions (positive or negative). In addition, the computer demonstrated either expert- or novice-level performance.
Since task partitioning presumably relieves the operator of performing the entire task, overall better performance should have been observed in the partitioned mode. However, computer skill level was predicted to interact with task mode, and participants in the expert condition, partitioned mode were expected to exhibit significantly better performance than those in the novice condition, partitioned mode. This would show the benefits of having an expert-level computer as a teammate and would coincide with Scerbo et al.