Does One Bad Machine Spoil the Bunch?: A Review of Trust in Multiple-Component Systems

Abstract

Human-automation interactions are rapidly transitioning from single-component automated systems to multiple-component systems. The human-automation literature has yet to adequately explore trust within multiple-component systems. A currently unanswered question is whether one faulty component causes an operator to lose trust in that one component (Component-Specific Trust; CST) or in every component in the system (System-Wide Trust; SWT). The goals of this paper were to 1) summarize the current work on trust in multiple-component systems, and 2) identify any trends that emerge during the literature review. We reviewed 17 experimental studies that tested whether operators tend to adopt CST or SWT under different conditions. Overall, most studies suggest that operators adopt SWT. However, studies that provided the operator with high decisional freedom and more time with the automated systems suggest that CST is the dominant strategy. Future work should explicitly test these and other variables that may promote users to adopt CST.

System-Wide Trust, Graded System Distrust, or simple Anchoring/Priming Effects?

“…While the study of factors that influence trust is prominent in trust in automation literature, little has been done to examine how trust is targeted in the first place. Lee and See (2004) provided the concept of trust specificity-the degree to which trust is discriminated towards individual components-but comparatively little work has actively explored trust specificity (Geels-Blair et al, 2013;Lopez & Pak, 2020). Even now, there are many important unresolved questions regarding how individuals form trust in complex systems.…”

“…For example, failure of a Tesla's autoparking feature may reduce trust in that Tesla's ability to safely use the 'Autopilot' feature. The alternative to the use of the SWT strategy is the component-specific trust (CST) strategy wherein trust is discriminated between individual components (Lopez & Pak, 2020). In the Tesla example above, this would result in no impact on trust in the 'Autopilot' feature following a failure of the autoparking feature.…”

“…In the Tesla example above, this would result in no impact on trust in the 'Autopilot' feature following a failure of the autoparking feature. Trust tends to be more calibrated when the CST assessment strategy is used, the use of a SWT strategy is generally more frequent in the literature (Lopez & Pak, 2020). The fact that more information regarding a system can lead to greater use of a CST strategy, but not fully mitigate an SWT contagion effect, makes further research of the phenomena important (Rice & Geels-Blair, 2010).…”

How Do Advanced Driver Assistance Systems Fit into Level of Automation Frameworks?

“…Alternatively, CST is more likely to occur when a user has more time and experience with an automated multicomponent system. According to a recent literature review, however, cases of SWT seem to be more prevalent in the literature (Lopez & Pak, 2020). The greater amount of time and experience needed to develop CST and the sufficiently refined mental models it relies on may not be established if disuse occurs because of the poor ratings earned by a lesser ADAS system and the poor SWT that may result.…”

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

“…The goal for designers is to develop systems that facilitate calibrated trust wherein an individual's trust in the automation and the automation's capabilities correspond (Lee & See, 2004, Merritt et al, 2015. However, this may prove difficult moving forward; with the increasing complexity of automation, many systems will likely consist of an array of subsystems or components (Atchley et al, 2022;Lopez & Pak, 2020).…”