Carl Westin

172 Discussion and recommendations Methods can be found in other fields such as recommender systems 201 and user- adaptive automation. 193 Personalization can, for example, be facilitated by the cre- ation of individual user profiles generated from an individual’s system-interaction data, alternatively data from other operators with similar preferences. For instance, Regtuit proposed a machine-learning technique consisting of k-means clustering for identifying consistent patterns in controllers conflict solving strategies that can be used to develop a conformal conflict resolution decision aid. 261 Moreover, physio- logical measures can be used to, for example, collect operators’ eye-gaze patterns in real time for purposes of dynamically track their behavior and adapt the visualiza- tion to their visual and cognitive abilities, needs, and preferences. 288 Personalization can be assessed together with neuroscience, by studying how individual differences in problem-solving and cognitive styles are associated with the activation of cogni- tive processes in different brain areas. 198 Accommodating individual problem-solving in EID. As an alternative to tai- lored decision advisories, a system can accommodate personalization by allowing a person to solve problems in their preferred way. In many contexts, including ATC CD&R, this approach may be preferable to that of a decision aid suggesting a spe- cific solution. For instance, ecological interfaces, such as the SSD, can visualize the constraints and space of solution possibilities affecting a situation and problem, but leave the final decision-making to the operator. This acknowledges that humans can be heterogeneous (the subjective truth), while situations and the constraints are homogeneous (the objective truth). In addition, another strength of this approach is that advisory timing does not need to be considered. This approach for facilitating personalization is particularly relevant to work domains in which ample time is available for solving problems. However, in time- critical and complex domains, such as ATC CD&R, problem-solving time is often limited. Furthermore, less time for solving conflicts can be expected as more tac- tical and sophisticated automation is introduced for purposes of increasing traffic throughput and efficiency. In light of these changes, the controller’s role can be ex- pected to change from tactical traffic intervention to supervising CD&R automation and only infrequently solve specific conflicts. As a consequence, the relevance of conformal automation may increase. 7-4-10 Side effects and ethical concerns The reliability of advisories was not varied. Although controllers sometimes dis- agreed with advisories, they may not have had enough reason to reject advisories and therefore accepted more. A question is whether unreliable conformal automa- tion affects acceptance and trust differently from other types of automation that are