The More You Know: Trust Dynamics and Calibration in Highly Automated Driving and the Effects of Take-Overs, System Malfunction, and System Transparency

Abstract

<jats:sec><jats:title>Objective</jats:title><jats:p> This paper presents a theoretical model and two simulator studies on the psychological processes during early trust calibration in automated vehicles. </jats:p></jats:sec><jats:sec><jats:title>Background</jats:title><jats:p> The positive outcomes of automation can only reach their full potential if a calibrated level of trust is achieved. In this process, information on system capabilities and limitations plays a crucial role. </jats:p></jats:sec><jats:sec><jats:title>Method</jats:title><jats:p> In two simulator experiments, trust was repeatedly measured during an automated drive. In Study 1, all participants in a two-group experiment experienced a system-initiated take-over, and the occurrence of a system malfunction was manipulated. In Study 2 in a 2 × 2 between-subject design, system transparency was manipulated as an additional factor. </jats:p></jats:sec><jats:sec><jats:title>Results</jats:title><jats:p> Trust was found to increase during the first interactions progressively. In Study 1, take-overs led to a temporary decrease in trust, as did malfunctions in both studies. Interestingly, trust was reestablished in the course of interaction for take-overs and malfunctions. In Study 2, the high transparency condition did not show a temporary decline in trust after a malfunction. </jats:p></jats:sec><jats:sec><jats:title>Conclusion</jats:title><jats:p> Trust is calibrated along provided information prior to and during the initial drive with an automated vehicle. The experience of take-overs and malfunctions leads to a temporary decline in trust that was recovered in the course of error-free interaction. The temporary decrease can be prevented by providing transparent information prior to system interaction. </jats:p></jats:sec><jats:sec><jats:title>Application</jats:title><jats:p> Transparency, also about potential limitations of the system, plays an important role in this process and should be considered in the design of tutorials and human-machine interaction (HMI) concepts of automated vehicles. </jats:p></jats:sec>

Scared to Trust? – Predicting Trust in Highly Automated Driving by Depressiveness, Negative Self-Evaluations and State Anxiety

“…This research enhances the understanding of emotional processes at play in the phase of getting to know and understand a newly introduced automated system. The moderate to high effect size of the relationship between anxiety and trust in automation, explaining 16% of the variance in trust, further highlights the importance of internal situational factors, as proposed in trust in automation models (e.g., Hoff and Bashir, 2015;Kraus et al, 2019b). The role of anxiety for the formation of trust toward a new and unfamiliar automated system underlines the affective facet of trust in automation (e.g., Lee and See, 2004).…”

“…If they indicated to not feel safe, they could continue the practice for a while, otherwise the study procedure continued. In the instruction text and in the practice trial, transparency of the automation was manipulated in two level (low vs. high transparency), which was not in the scope of this research (Kraus et al, 2019b). Afterwards, participants were instructed that they would now drive with the automated driving system; they had been introduced to earlier.…”

“…State anxiety was measured with a translated short version of the State-Trait Anxiety Inventory (STAI-6, Marteau and Bekker, 1992), in which participants had to assess their current feelings with six statements. Trust in automation was assessed with a German shortened version of the Automation Trust Scale (ATS; Jian et al, 2000) with seven items (see also Kraus et al, 2019b). The scale was designed to provide an economic measurement of trust in automation as a unidimensional construct.…”

“…Conversely, if users' trust is not calibrated, the dangers of disuse (underachieving the full potential of a system) and misuse (overstretching the capabilities of a system) of the system are increased (Parasuraman and Riley, 1997). To prevent these negative outcomes and in order to facilitate trust calibration, a thorough understanding of the psychological processes associated with trust formation and calibration is essential (e.g., Lee and See, 2004;Hoff and Bashir, 2015;Kraus et al, 2019b). Especially the relation of personality traits and emotional states to individual differences in trust in automation are a promising direction to gain an understanding for the underlying psychological mechanisms of trust formation and calibration (e.g., Hoff and Bashir, 2015).…”

More Than a Feeling—Interrelation of Trust Layers in Human-Robot Interaction and the Role of User Dispositions and State Anxiety

“…Within learned trust, one can further distinguish between initial learned trust based on information and existing knowledge prior to the interaction with a system and dynamic learned trust, which refers to trust adaptions during the actual interaction with a given system. It follows that learned trust is subject to change over time and is updated before and during the interaction by accumulated information and observations, for example, on perceived system performance (Merritt and Ilgen, 2008;Kraus et al, 2019b). It is further assumed that this process of learning to trust follows to a considerable extent the mechanisms of attitude formation and change (see Maio et al, 2018, exemplarily).…”