Missing key information: How automation failure can be misinterpreted

Abstract

Human monitoring of systems in which sensors provide data to automated decision support algorithms create interesting challenges for Human Factors. In this study we are interested in exploring whether people are able to detect two types of automation failure: when decisions do not fit the data presented to the operator, and when data from different information sources do not agree. For those students that performed at a level of ≥ 97% correct ('high performers), checking for both types of failure seemed easy. For those students that performed at a level of ≤ 95% correct ('low performers'), checking for erroneous recommendations seems straightforward, but checking for information agreement seemed to be omitted. One suggestion is that the non-experts expended more effort on checking recommendation and ignored the need to check congruence across displays. The implication is that the 'worth' of the displayed information for one task (decision check) outweighed its worth for the simpler task (congruence check) for the non-experts.