RECENT PRESENTATIONS

Gevins, A., & Smith, M.E. (2005). Neurophysiologic measures for neuroergonomics. HCI International 2005, July, Las Vegas, NV.

ABSTRACT

It is a compelling, futuristic idea to use brain activation measures as a basis for automated systems to off-load tasks from an individual if he or she was in a state of high cognitive workload, or allocate more tasks to an individual that appeared to have ample reserve processing capacity. Like many great visions, the devil is in the details. In this presentation we describe recent progress in the development of multivariate neurophysiologic metrics of regional cortical brain activation. In a first experiment, EEG recordings were made during a daytime session while 16 well-rested participants performed versions of a PC flight simulator task that were either low, moderate, or high in difficulty. In a second experiment, the same subjects repeatedly performed high difficulty versions of the same task during an all night session with total sleep deprivation. Multivariate EEG metrics of cortical brain activation were computed for frontal brain regions essential for working memory and executive control processes crucial for maintaining situational awareness, central brain regions essential for sensorimotor control, and posterior parietal and occipital regions essential for visuoperceptual processing. During the daytime session each of these regional measures displayed greater activation during the high difficulty task than during the low difficulty task, and degree of cortical activation correlated positively with subjective workload ratings in these well-rested subjects. During the overnight session, cortical activation declined with time-on-task, and the degree of this decline over frontal regions correlated negatively with subjective workload ratings. Since participants were already highly skilled in the task, such changes likely reflect fatigue-related diminishment of frontal executive capability rather than practice effects. In a third experiment, a prototype online system for concurrently deriving regional indices of cortical brain activation was implemented and tested. The prototype was validated empirically by using it to gauge brain function in ten well-rested participants while workload was varied through task difficulty manipulations. The results confirmed that cortical brain activation indices derived in real time significantly differed between relatively easy and relatively difficult tasks. As in the first experiment, frontal cortical activation correlated positively with subjective workload. The current results thus indicate that a decrease in cortical activation in frontal regions may either reflect a decrease in mental workload, or an increase in mental fatigue and a heightened sense of mental stress. This is problematic for the development of brain-adaptive automation systems. Assigning more tasks to an individual in the former case may indeed serve to increase his or her cognitive throughput, but in the latter case it could be quite counterproductive, even resulting in the sort of tragic accident that too often occurs when fatigued personnel are confronted with unexpected increases in task demands. For such neuroadaptive systems to successfully modulate the cognitive task demands placed on an individual in response to momentary variations in the availability of mental resources as indexed by neural activation measures, sophisticated user models with human-like intelligence will need to be developed that take both task demands and the operator's state of alertness into account. It is realistic to expect that a great deal of methodical research will be needed before brain activation measures can actually be put to use for adaptively augmenting the capabilities of mission-critical personnel working at demanding and stressful tasks. Further systematic developments such as future progress with the measurement approach outlined here will undoubtedly eventually lead to an effective and practical means for monitoring transient changes in cognitive brain function that is prerequisite to actual neuroadaptive automation. Supported by DARPA.