Sustaining attention under monitoring conditions: What changes in the brain when attention lapses?

The move to semi-automated systems has created many situations where humans are required to sustain attention over time with only infrequent responses (e.g., monitoring in train and aircraft network control, semi-automated vehicles). Unfortunately, when targets are rare, observers are more likely to miss them or be slow to react (e.g., Wolfe et al. 2005). Our Multiple Object Monitoring (MOM) task includes many key features of modern environments: it requires selection of targets from distractors, tracking of dynamic stimuli, and allows distinction of time-on-task effects from target frequency effects. In the MOM task, dots approach a central obstacle from different directions along predictable trajectories. The task is to avoid a collision between dots of the relevant colour and the central obstacle. On a variable proportion of trials, the dots are automatically deflected away from collision at a set point in the trajectory. In our original study, we found that in ‘active’ conditions, where participants manually deflect 50% of dots; they can successfully sustain performance across time. In monitoring conditions, where participants manually deflected 6% of dots, performance dropped sharply over time, showing a clear vigilance decrement. Here, we report task parameter manipulations that show vigilance decrements decrease with increasing target frequency, as well as using electroencephalography (EEG) combined with multivariate pattern analyses (MVPA) to examine patterns of neural activity during successful and lapsed attention. Our neural studies show that we can use the pattern of activity across the brain to identify information about task-relevant aspects of the task (e.g., target dot location) and use the neural difference between correct and miss trials to predict behavioural results. These findings contribute to our understanding of successful versus unsuccessful sustained attention and provide a foundation for predicting behavioural errors before they occur, based on neural patterns of activation.