User trust here and now but not necessarily there and then - A Design Perspective on Appropriate Trust in Automated Vehicles (AVs)

Doctoral thesis, 2023

Automation may carry out functions previously conducted only by humans. In the past, interaction with automation was primarily designed for, and used by, users with special training (pilots in aviation or operators in the process industry for example) but since automation has developed and matured, it has also become more available to users who have no additional training on automation such as users of automated vehicles (AVs). However, before we can reap the benefits of AV use, users must first trust the vehicles. According to earlier studies on trust in automation (TiA), user trust is a precondition for the use of automated systems not only because it is essential to user acceptance, but also because it is a prerequisite for a good user experience. Furthermore, that user trust is appropriate in relation to the actual performance of the AV, that is, user trust is calibrated to the capabilities and limitations of the AV. Otherwise, it may lead to misuse or disuse of the AV.
     The issue of how to design for appropriate user trust was approached from a user-centred design perspective based on earlier TiA theories and was addressed in four user studies using mixed-method research designs. The four studies involved three types of AVs; an automated car, an automated public transport bus as well as an automated delivery bot for last-mile deliveries (LMD) of parcels. The users’ ranged from ordinary car drivers, bus drivers, public transport commuters and logistic personnel.
     The findings show that user trust in the AVs was primarily affected by information relating to the performance of the AV. That is factors such as, how predictable, reliable and capable the AV was perceived to be conducting for instance a task, as well as how appropriate the behaviour of the AV was perceived to be for conducting the task and whether or not the user understood why the AV behaved as it did when conducting the task. Secondly, it was also found that contextual aspects influenced user trust in AVs. This primarily related to the users’ perception of risk for oneself and others as well as perceptions of task difficulty. That is, user trust was affected by the perception of risk for oneself but also by the possible risks the AV could impose on other e.g. road users. The perception of task difficulty influenced user trust in situations when a task was perceived as (too) easy, the user could not judge the trustworthiness of the AV or when the AV increased the task difficulty for the user thus adding to negative outcomes. Therefore, AV-related trust factors and contextual aspects are important to consider when designing for appropriate user trust in different types of AVs operating in different domains.
     However, from a more in-depth cross-study analysis and consequent synthesis it was found that when designing for appropriate user trust the earlier mentioned factors and aspects should be considered but should not be the focus. They are effects, that is the user’s interpretation of information originating from the behaviour of the AV in a particular context which in turn are the consequence of the following design variables: (I) The Who i.e. the AV, (II) What the AV does, (III) by What Means the AV does something, (IV) When the AV does something, (V) Why the AV does something and(VI) Where the AV does something, as well as the interplay between them. Furthermore, it was found that user trust is affected by the interdependency between (II) What the AV does and (VI) Where the AV does something; this was always assessed together by the user in turn affecting user trust. From these findings a tentative Framework of Trust Analysis & Design was developed. The framework can be used as a ‘tool-for-thought’ and accounts for the activity conducted by the AV, the context as well as their interdependence that ultimately affect user trust.