Abstract
From self-learning thermostats to agricultural vehicles to baggage-screening systems, automation is becoming an integral part of modern life. In safety-critical domains, automation offers incredible benefits by improving the accuracy of complex tasks. Because automation failures are inevitable, appropriate calibration of operator trust in automation is integral for the optimal performance of human–automation teams.
Trust in automation is affected by several factors, such as automation reliability, operator workload and self-confidence, task difficulty and novelty, consequences of automation misses, and automation display format and content. Trust is therefore a multidimensional concept. So what can be done to appropriately calibrate operator trust level on automation?
Train operators perform differently based on their beliefs about automation (Pop, Shrewsbury, & Durso, 2015). For example, operators with low expectation that automation is trustworthy do not comprehend the situations in which automation differs in reliability and should be given information around the external factors contributing to automation errors so as to develop appropriate trust in automation.
Provide contextual information about the vulnerabilities of the automation algorithm to help operators overcome automation misses under ambiguous circumstances (Rovira, Cross, Leitch, & Bonaceto, 2014). This information is especially useful during high-workload situations, when the reliance on automation is higher compared with low-workload situations.
To minimize operator errors, decision aids should indicate the lower degree of confidence along with predictive advice during uncertain scenarios (Wickens, Clegg, Vieane, & Sebok, 2015).
Increase the anthropomorphism and politeness of the automation to make it trustworthy (Hoff & Bashir, 2015).
To contribute to the effective design of future automated systems that encourage appropriate trust, leverage human factors techniques, such as eye tracking and operator interviews, to understand expert operators’ visual attention patterns when using existing automated systems (Sanchez & Duncan, 2009).