Carl Westin

64 Source bias effects 4-2-1 Differences in trust between human and automated sources Although trust and acceptance in both automated aids and other humans have been researched extensively, differences between the two have been examined by only a few studies within the past twenty years. These indicate that theories of interper- sonal trust and acceptance partly apply to automated sources, 72, 160 although funda- mental differences exist. 48 In general, the decision to use automation for a specific task is considered to be driven by a core process of comparing the perceived trust, or confidence, in automation with the confidence in one’s own ability for accomplishing that task. 15, 50, 68, 108, 160–164 Automation is preferred when self-confidence is lower than trust in automation, while manual performance is preferred when trust in automa- tion is lower than self-confidence. 68, 71 Research suggest that a similar process takes place in relation to human sources. 50, 160, 164 In contrast to human sources, however, automated sources tend to be perceived as experts and assigned higher performance expectations prior interaction. 164, 165 Provided the automated source is perceived to perform well, high trust is main- tained and will likely result in high reliance on the aid. 166 However, if an aid errs, trust and acceptance will drop. Basically, the more the aid was trusted before erring, the more severe the trust reduction will be. This represents a key difference between trust and acceptance in automated and human sources. Notably, people seem more sensitive and critical to errors made by an automated sources than errors made by a human source. 165, 167 When acceptance and trust have been lost in an automated source, it takes time to regain during which period the automated source’s reliability is likely to be underestimated. In contrast, trust in human sources is less drastically affected by errors and more quickly repaired. Misuse is typically only discussed in relation to automated sources. It includes the overlapping concepts of perfect automation schema , automation bias , and au- tomation complacency . While these have been found to stem from similar atten- tional processes leading to overreliance in a system, they manifest slightly differ- ently. The perfect automation schema refers to exceedingly high expectations of au- tomation performance before use, which can cause disproportional large trust losses when these expectations are not met. Automation bias has been linked to errors of commission before and during use (depending on automation when it is wrong or performing badly) while complacency has been linked to errors of omission (failure of being vigilant and supervising the automation). 63, 168 In relation to these issues of overreliance, researchers have cautioned against the development of aggressive and confident decision aids in the face of uncertainties. 169 Different source biases are believe to be intertwined with the concepts of ex- pertise and pedigree. In a series of studies, Madhavan and Wiegmann explored the