Carl Westin

108 Automation transparency effects Secondly, results supported the hypothesis that increased automation trans- parency can be achieved by increasing the amount of visualized meta-information. Questionnaire results indicated an overall preference for the more transparent TRI SSD, which correspond to previous research results in transparency research that the system and its advisories are better understood when more meta-information is provided about the criteria affecting problem-solving. 191, 195 Further, in accordance with previous research, questionnaire results showed that increased transparency was associated with perceptual costs in terms of increased clutter and working dif- ficulties. These perceptual differences did, however, not translate into significant differences in the simulation data. While the current study was limited by a small sample size, results suggest that although transparency was perceived as an impor- tant characteristic of automation, it was less relevant for the actual interaction with an interface. Similarly, Sadler et al. found that acceptance of aircraft diversion recommendations were not affected by transparency, although higher transparency in their study was significantly associated with higher trust ratings in the decision support system. 195 Future research should further investigate to what extent the per- ception of transparency influences the acceptance and interaction with automated systems. 5-7-1 Calibrating transparency This study has shown that EID-derived support tools, such as the SSD, can success- fully be used to visualize meta-information in ways that facilitate increased trans- parency. The framework is especially suitable for deciding what to visualize and, to some extent, how to visualize it. However, the design framework does not specify how much information, or the extent of detail, that should be incorporated in the visualization. As such, EID does not provide much guidance for finding the appro- priate transparency balance between insufficient and excessive information. Further research should investigate ways to integrate guidelines in EID for establishing a degree of transparency that is appropriate given the task, context, and operator. Taken together, findings emphasize the need for calibrating transparency in or- der to facilitate appropriate understanding. Furthermore, although there is a strong incentive for reducing the amount of information presented to operators, there is an opposite intrinsic desire for the operator to process as much information as possible in order to make the best decision. While a shortage of information may not facil- itate the needed understanding, excessive information can overload the user. This relationship between understanding and meta-information can explain the mixed re- sults obtained in transparency related research, in which transparent systems have been shown to affect human-machine interaction and task performance both posi- tively and negatively.