Carl Westin

152 Discussion and recommendations mation was provided in an argument for the solution. Although the SSD interface provided an overview of the constraints affecting solutions, it did not grade or value options in any way. While conformal advisories received significantly higher agreement ratings in the First empirical study, the effect was not repeated in subsequent studies. A slightly reversed effect was noted in the Automation transparency study. The higher agreement for nonconformal advisories can be explained by the more transparent triangle SSD, which revealed more of the available solution space and made al- ternative solutions available that controllers had not thought of during the baseline prequel simulation (for which the heading band SSD was used). Controllers in the First empirical study responded faster to conformal advi- sories than nonconformal advisories (regardless of accept/reject decision). They also accepted conformal advisories faster than they accepted nonconformal advi- sories. This suggesting that their choice to accept advisories was deliberate and not due to a “blind” acceptance strategy driven by attitudes of trust or the perceived credibility of the system. Results were not reproduced in the Source bias and Au- tomation transparency studies. Although, a notable finding in the latter, was the faster response time to nonconformal advisories in the heading band condition as opposed to the faster response time to conformal advisories in the triangle condi- tion. A possible explanation is that controllers were trainees (although in their final educational stage), and had not developed strong preferences for solving conflicts. 7-1-2 Other factors affecting acceptance In addition to strategic conformance, several other factors influencing acceptance have been considered in this thesis. The following are believed to be particularly important and relevant for future sophisticated automated systems assuming more cognitive work. Level of automation and complexity. The First empirical study explored the interaction between strategic conformance together with the level of automation (LOA) and complexity on advisory acceptance. Two LOAs were used: management by consent (MbC), for which the automation suggests and operator authorizes, and management by exception (MbE), for which the automation acts autonomously but informs the operator who can intervene. Because of the subtle difference between LOA conditions, they were ultimately collapsed and not considered for analysis. This led to numerous discussions sur- rounding the LOA concept, leading to a consensus that the contribution of these frameworks is limited for guiding research and development. Similar criticism has been promulgated by artificial intelligence researchers, who argue for the abandon-