This research paper describes observed effects that automated procedural and decision aids have on human error rate. The hypothesis was that pilot crews tend to over-rely on automation to perform tasks and make decisions, instead of using automation as one component of the decision making process. Automation bias is analyzed by examining errors made as a result of automation task demand (automation commission errors), and errors that result from inadequate advisory by the automated system (automation omission errors). The purpose of this study was to investigate the relationship between cognitive bias, error rate, and criteria of imposed accountability for performance.

A test group was asked to perform a computer simulated tracking task. Each simulation included an Automated Monitoring Aid (AMA) that provided directives relating to task/performance procedures. Besides the AMA, each simulation was equipped with supplemental advisory systems for use in verification of the AMA. Participants where informed that the AMA was not 100% accurate. Reliability of AMA information was pseudo-randomly controlled. Test subjects were assigned to known groups of non-accountability and accountability. The results indicated the following: 1) subjects in the unreliable simulation performed better than the AMA as a result of verification; 2) subjects in the reliable simulation performed slightly better than by performing with no verification; and, as a surprise to the researchers, 3) all groups decreased their verification efforts as the demands of the verification task increased.

The conclusions offered by this study should lead specifically to improved safety and performance by pilot crews operating within the framework of accountability. This can be accomplished by better system design matching the cognitive needs of the end user performing within a framework of accountability. Of prime consideration is the need for additional research for improved design of systems structure and feedback as applied to workload within human-computer dialogues.