Carl Westin

22 Strategic Conformance: A Literature Survey ing fully automated concepts, such as AERA 1-3, 74, 75 ARC 2000, 76 and the PHARE project. 77 Realising that full automation was not feasible, and that the human has to retain control, several sub-concepts were instead transformed into decision aids. 10, 77 Their success in adoption, however, has been questioned. 9 Conflict-solving algorithms typically approach the environment in a dichoto- mous fashion, providing single fixed, optimal solutions according to causal de- terministic laws. 19 Most models are typically limited in application and options considered for conflict resolution. 11, 78 According to 19 all CD&R systems follow roughly the same design model, containing a deterministic trajectory model and set of alerting threshold metrics, often based on “engineering intuition.” After a series of simulations threshold alerts are tuned for optimal performance. Finally, perfor- mance is assessed in terms of false alarm rates and loss of separation frequency. In complex decision-making tasks, operators do not necessarily agree with their decision aid on when/whether to intervene, nor which solution to apply. As such, the decision aid can be seen to possess its own implicit decision-making style that is either conformal with the operator’s or not. For example, in recent real-time ATC simulations exploring novel decision aids, the automation was perceived as occa- sionally “fighting” against the controller on how to solve conflicts. 17 In another study, the hampered adoption of a conflict detection aid (URET, or User Request Evaluation Tool), was attributed to the system not reflecting controllers current con- flict managing procedures. Controllers felt that the aid was too slow (its advice often coming after conflict detection) and, contrary to the designers’ intentions, tended to use the tool instead as a flight strip replacement and route amendment tool. 9 Deciding on when, and on which thresholds, automation should interact may be one of the most difficult problems to address, and constitutes perhaps the most pressing mismatch between controller and automation in ATC. 61 Additionally, there might be large individual differences in preferred safety margins. The decision of whether to intervene can generally be described as a trade-off between intervening directly or waiting and collecting more information. 56 Both choices can be deter- mined by a workload regulation process. 79, 80 Intervening directly can be appealing to avoid the additional workload of having to monitor the evolving situation. In busy situations, however, task shedding and prioritizing may encourage postponing intervention in order to control and maintain minimum workload. In certain situ- ations, neither early nor late intervention may be feasible, and can put controllers in a “double bind.” CD&R automation, however, is generally not this sensitive, but operates according to pre-defined rules and process parameters. 56