Carl Westin

166 Discussion and recommendations arguments reflect two questions: will automation thinking and making decisions like a human not make the same mistakes as the human? And, what is the bene- fit of automation making the same decisions as the human would? Although these questions touch upon two weaknesses of strategic conformance, they have, to some extent, been addressed in previous research. In addressing the first weakness, a possible approach is to identify and elim- inate poor decision-making strategies and focus on “best practice” strategies (e.g., CORA). A similar approach is advocated by Gigerenzer and colleagues 20, 82, 268 who argue that heuristics can be quantified into formal models (called fast-and-frugal decision trees) that not only are more compatible with human decision-making, but also outperform other logical and statistical models in accuracy. A similar, slightly more advanced approach can be found in bootstrapping models, which are a type of expert system based on the quantified judgment and decision-making rules and strategies made by experts. Bootstrapping models are thought to improve the re- liability and accuracy of judgments, since an expert’s consideration of a criterion, and how it is weighed, is applied consistently. On the other hand, bootstrapping models lack the flexibility and adaptability of human experts, and cannot consider variables or cues outside the scope of the model. As with fast-and-frugal heuristics, the accuracy of bootstrapping models often deteriorates when more variables are considered. 277 Similar to heuristic and bootstrapping models, strategic conformal automation would embrace the “good and safe” practices. A CD&R decision aid, for example, would be attuned to individual workload demands, and allow situations to progress longer during low workload episodes to minimize path deviations and promote ef- ficiency. Conversely, during high workload situations, automation could intervene early and apply larger separation thresholds, the strategy being to “set and forget” 24 and allow the controller to move on to the next task. Other appropriate controller heuristics include: • approach conflicts in pairs and sequentially, rather than globally, determine if multiple aircraft are involved; 35 • avoid interfering with both conflicting aircraft in lower airspace; 88 • being conservative in conflict detection, identifying a surplus of potential con- flicts; 87 • a tendency to operate at larger separation thresholds than is strictly re- quired; 61, 98, 278 and • safer to turn slower aircraft behind faster; 24, 35