Systems with Human Monitors: A Signal Detection Analysis

Abstract Automated factories, the flightdecks of commercial aircraft, and the control rooms of power plants are examples of decision-making environments in which a human operator performs an alerted-monitor role. These human-machine systems include automated monitor or alerting subsystems operating in support of a human monitor. The automated monitor subsystem makes preprogrammed decisions about the state of the underlying process based on current inputs and expectations about normal/abnormal operating conditions. When alerted by the automated monitor subsystem, the human monitor may analyze input data, confirm or disconfirm the decision made by the automated monitor, and take appropriate further action. In this paper, the combined automated monitor-human monitor system is modeled as a signal detection system in which the human operator and the automated component monitor partially correlated noisy channels. The signal detection analysis shows that overall system performance is highly sensitive to the interaction between the human's monitoring strategy and the decision parameter, Ca, of the automated monitor subsystem. Usual design practice is to set Ca to a value that optimizes the automated monitor's detection and false alarm rates. Our analysis shows that this setting will not yield optimal performance for the overall human-machine system. Furthermore, overall system performance may be limited to a narrow range of realizable detection and error rates. As a result, large gains in system performance can be achieved by manipulating the parameters of the automated monitor subsystem in light of the workload characteristics of the human operator.