How to support trust in complex system

One of the major factors for safe and efficient utilisation of new automated complex systems is the level of trust that an operator has in the system.;This study encompasses two objectives. The first one is to propose design rules and guidelines for system designers on how to support an appropriate level of trust in automated systems, in particular in novel systems. The proposed guidelines are based on extensive theoretical, experimental and empirical results on trust in automated systems available in the research literature.;The second objective is to verify the impact of interface usability on the level of trust in the system. We hypothesized that the enhancement of the interface usability supports trust. In order to test this hypothesis the usability of a new remote sensing application IDFS (Intelligent Data Fusion System) was evaluated by two methods, one based on heuristics (conducted by the author of this study) and the other based on a walkthrough (conducted with six expert users in the domain of remote sensing) methods. During the evaluation a particular importance was paid to the issues of trust. We developed recommendations on possible improvements of usability of the IDFS interface aimed at trust enhancement. Then these recommendations were implemented in the modified version of IDFS. In particular, the compatibility of the system's behavior with the user's expectations, informative feedback, visual clarity and traceability of the information have been enhanced.;Both the new (redesigned) and the old (initial) interfaces of the IDFS have been tested by six users, namely, graduate students doing research in the domain of remote sensing at the University of Montreal (Department of Geography). The level of trust was measured using the trust measurement scale of Jian et al. (2000). Even though the operators were well aware that the system functionalities in the two versions of the IDFS were mostly the same, the measurements revealed that the new interface did increase trust by a factor of two. Thus, these results support the hypothesis that usability promotes trust. They also are consistent with the proposition that in early interactions with a new system trust strongly depends on affective1 and analogical2 factors. It is thus concluded that enhancement of interface qualities such as compatibility of the system's behavior with the user's expectations, informative feedback, visual clarity and traceability, which all support development of the affective and analogical trust.;The test results also show that the scale of Jian et al. (2000) does allow the measurement of the initial trust in novel systems (i.e. when there is no prior experience with the system). However, the scale does not provide a description for the terminology that is used. We propose to explicitly explain each item of the scale in order to help the users understand the terminology. Another point is that the scale of Jian et al. (2000) consists of 12 answers for 12 different trust-related items without proposing unified single measure for the overall trust in the system.;In the study we built on the trust measurement scale of by Jian et al. (2000), and suggested a model that enables measurement of the overall trust. For both the old and the new interfaces the dispersion of individual items in the scale from the overall trust value is small: sigmaold≈0.077 and sigmanew≈0.076, respectively. This reveals that the individual items in this scale are very close to each other or they measure aspects of trust that is hard for the subjects to differentiate.;The limitations of experimental results and the trust measurement model we have developed are also discussed. It would have been preferable to measure trust with a larger number of users, tasks, and functionalities in the system.;Future research is necessary to reveal these differences. For example, larger dispersion could be an indicating for a larger confusion in the meaning of item. However, dispersion cannot be the single criteria for assigning the weight. Other factors should be taken into account, such as direct bearing of that each item to trust.;Another future research avenue is to find out the impact of other usability qualities such as help and guidance, consistency, error prevention, flexibility and rapidity of use on trust. One way to measure the impact of different interface qualities on trust is to design an experiment with two versions of interface: one with the interface qualities and the other without them. The comparison of these results will show which interface qualities play major role in trust development.;1 Affective trust tuning involves essentially anything which results in producing positive or negative feelings toward the system. 2 Analogical trust is developed by comparing the system behavior with rule-based expectations. When rules are consistent with trustworthy behavior, they can increase people's expectations of satisfactory performance.