Carl Westin

90 Automation transparency effects ibrated among drivers when they were provided with a visualization of the cars ability to autonomously drive. Results indicate that drivers had a better understand- ing of the system’s ability and its limitations. They also intervened faster and more appropriately when the autonomous car reached its operational limits and suddenly failed. In another experiment, air defense operator’s understanding of a decision support system’s reasoning in target detection increased as a result of indicating the system’s limitations by means of uncertainty estimates for detected targets. In addition, operators became more attentive and cautious in their identifying targets. However, the effects of increasing transparency have not been uniformly posi- tive. Drawing on mixed results from the same experimental series, Helldin 191 noted that both workload and decision-making time suffered (i.e., increased) with increas- ing transparency. 191 Cramer et al. 111 found that explaining to users why an advisory was made increased acceptance but did not influence their trust in the system. Our previous research suggests that inexperienced operators’ acceptance of (accept or reject), and agreement with (on a 1-100 rating scale), solution advisories is not in- fluenced by decision aid transparency in regards to the visualization of constraints affecting conflict resolution advisories in an ATC CD&R. 217 Mixed results have also been attained in research by Bass and colleagues, who found that risk probability judgments in an egocentric aircraft conflict prediction task did not improve with decision aid transparency achieved by accompanying probability estimates with strategy information (i.e., a visualization of the closest point of approach between the two aircraft). Participants’ poor performance in the more transparent condition was attributed to the interface representation being too confusing, showing probability contours for risk of conflict, in the combination with insufficient training. 218 While more recent study reversed these findings (i.e., prob- ability judgments improved with strategy information), the provision of strategy information was found to be less effective than the provision of uncertainty infor- mation pertaining to the speed and heading of other aircraft (referred to as environ- mental information). 219 Overall, these empirical findings do not provide a coherent picture of automa- tion transparency and how it affects human attitudes and automation use behavior. However, the seemingly contradictory outcomes of transparency research can pos- sibly be explained by an insufficient balance of information. In all studies reviewed herein, increased transparency is consistently achieved by providing more meta- information pertaining to, for example, system reasoning, 111, 191, 195, 196 information uncertainty, 191, 218, 219 or situation awareness. 200, 216