System-Wide Trust: The Impact of an Error in a Multi-component System

Abstract

Given the rise of automation adoption in a variety of industries, it is essential we understand how individuals perceive these systems. Previous studies have found that a failure of one component in a system leads to decreased trust across a whole system, but few, if any, studies have considered how users define the confines of a system. To address this gap, this study replicated and extended Mehta et al. (2019) by incorporating measures of trust, similarity, and functional relatedness between six human crew members and six automated components. We noted that our functional relatedness and similarity measures were predictors of the magnitude of trust difference between conditions, with a high degree of shared variance. Our lack of a within subjects design means that future studies should further test whether measures approximating perceptions of ‘system’ (e.g., perceived functional relatedness and similarity) predict the size of contagion effects.

How do Blame Attributions Impact Trust in Complex Task Environments?

“…Our study supported that the share of blame, but not responsibility, attributed to each component predicted the magnitude of contagion effects on trust. Echoing the findings of previous SWT research, this grading of the contagion effect suggests that this is not entirely an anchoring or response bias effect but an effect of contagion (O’Hear et al, 2022).…”

“…Doing so would have provided a more valid and reliable measure of trust but would also have been impractical, given the number of components used. Many previous studies on the topic display the same limitation (O'Hear et al, 2022;Mehta & Rice, 2016;Mehta et al, 2019). Future research exploring SWT contagion effects should consider using a more engaging, interactive, and longitudinal experimental paradigm and a validated measure of trust.…”

“…Previous findings have reported SWT ‘contagion effects’ such that a failure in one of a system’s components can be a predictor of a loss in trust across non-failing components of that system (Geels-Blair, Rice, and Schwark, 2013; Mehta et al, 2019). A recent study noted that similarity could be an important predictor of the size of contagion effects (O’Hear et al, 2022). No studies to date have considered how blame attributions predict the magnitude of contagion effects.…”

“…Said another way, it is unclear, from the previous studies, whether SWT effects are caused by true changes in user trust of the different components of the 'system', or if these SWT effects do not actually represent a change in trust at all but are rather caused by some non-trust related bias in the manner of response. Recent work points towards the possibility that SWT contagion effects are applied based on, at least in part, perceived similarity to the failing component (O'Hear et al, 2022). Future research is still needed to investigate the mechanisms by which contagion effects are applied, and to determine whether priming/anchoring effects can be a potential explanation.…”

“…For instance, researchers have found that SWT contagion effects extend toward paradigms utilizing gauges or UAVs, which are dissimilar in appearance (Wallister, de Visser, & Shaw, 2016; Bean, Rice, & Keller, 2011). A few recent studies have even noted that participants sometimes group humans and automated components for system-wide assessment, with a human being trusted less following automation errors and vice versa (Mehta & Rice, 2016; Mehta et al, 2019; O’Hear et al, 2022). Given this, user perceptions of what entails a ‘system’ may be quite inclusive.…”