Stereotypical of Us to Stereotype Them: The Effect of System-Wide Trust on Heterogeneous Populations of Unmanned Autonomous Vehicles

Abstract

Operators generalize their trust across all of the autonomous agents they are working with, a phenomenon referred to as System Wide Trust (SWT). As a result, the failure of one aid can cause a trust decrement in —and therefore disuse of— all other competent aids within the system. This study explored two possible SWT mitigation strategies: competence transparency and different appearance of aids. Previous research has shown that transparency and feedback affects trust calibration in systems (Walliser et al., 2016), yet our appearance manipulation is relatively novel, using gestalt principles of grouping to explore whether heterogeneous aids will be more easily differentiated compared to homogenous aids. Participants supervised four UAVs that identified targets as enemies or friendlies. Only one of these UAVs was inaccurate (70% recommended accuracy), which caused a SWT trust decrement for all UAVs. We expected that the heterogeneous UAVs with competence transparency would suffer the least SWT effect, yet the results did not find a difference between conditions. These findings suggest that the System Wide Trust effect is too strong to be affected by our manipulations and that further research on mitigation strategies is required.

Is the Pull-Down Effect Overstated? An Examination of Trust Propagation Among Fighter Pilots in a High-Fidelity Simulation

“…The present study offers several commonalities with prior research. Our experimental design included a comparable ratio of unreliable system components (i.e., 1:4) to reliable system components (i.e., 3:4) as in previous studies (Kluck et al, 2018; Lopez et al, 2023; Rice & Geels, 2010; Walliser et al, 2016). Though this study included a uniquely high-fidelity flight simulator, the task remained consistent with a few prior studies in using a simulated aviation task to monitor UAVs (Foroughi et al, 2019; Kluck et al, 2018; Walliser et al, 2016).…”

“…Our experimental design included a comparable ratio of unreliable system components (i.e., 1:4) to reliable system components (i.e., 3:4) as in previous studies (Kluck et al, 2018; Lopez et al, 2023; Rice & Geels, 2010; Walliser et al, 2016). Though this study included a uniquely high-fidelity flight simulator, the task remained consistent with a few prior studies in using a simulated aviation task to monitor UAVs (Foroughi et al, 2019; Kluck et al, 2018; Walliser et al, 2016). The use of a self-report measure of trust in automation to detect a pull-down effect is consist with the methods of Blair et al (2012), Kluck et al (2018), Walliser et al (2018), and Walliser et al (2023).…”

“…In contrast, previous studies that observed a pull-down effect of trust in automation relied almost exclusively on undergraduate samples and may have less generalizability to expert operators. Because the pulldown effect was not observed, it is also unclear how heterogeneity (e.g., differently named CCA platforms) might mitigate this form of trust bias; however, previous attempts to convey heterogeneity in system components (i.e., through distinct UAV symbology) have also failed to mitigate the pull-down effect (Bean et al, 2011;Kluck et al, 2018).…”

“…However, a subsequent study found that the pull-down effect persisted even when system components were designed with visually distinctive shapes (Bean et al, 2011). A related hypothesis was not supported by a flight simulation experiment that manipulated color, size, and wing shape of UAV indicators (Kluck et al, 2018). Research on this potential mitigating factor is scarce, leaving other means of heterogeneity (e.g., mixed-asset swarms) largely unexplored.…”

Measuring and Understanding Trust Calibrations for Automated Systems: A Survey of the State-Of-The-Art and Future Directions

“…Screening of dangerous objects/subjects [4, 14, 15, 19, 36, 51, 54-56, 85, 94, 107, 117], detection of system malfunctions [89,90], crime prevention [10], recidivism prediction [132], watch a video of a house search [66,83] Transportation Responding to take-over requests [2, 6, 7, 47, 61, 68-70, 77, 84, 98], collusion avoidance [8], managing (air) traffic [31,118], pedestrians interaction with AV [48], observe AVs [64,119], drive in driving simulator [87,96,136] Military Screening tasks [17,33,44,63,81,82,127,130,135,139,140], gathering of information [58], mission planning [92], human-AI collaboration for search and destroy missions [116] Production Improving production [73,133,143], disassembly [5], moving objects [34,35,45], demand forecasting [39], harvesting [113], quality checks [142] Gaming trust game [3,23], collaboration game [24], flanker task…”