Carl Westin

1-5 Research approach 7 not be able to observe or follow the reasoning process underlying a solution, these processes would be “apparent” as they only can be inferred from the observable automation behavior. As such, the goal of this thesis is as follows: Thesis goal To empirically investigate strategic conformance as a means for more personalized automation support, and develop a fundamental under- standing of how a decision aid’s strategic conformance affects the in- teraction with that aid and acceptance of its advisories. 1-5 Research approach To investigate this, a novel and ambitious research approach was developed based on a hybrid of methodologies, including state-of-the-art literature reviews, surveys, and several interrelated real-time simulations. Empirically investigating the con- cept of strategic conformance required a method for subjecting a controller to an automated resolution advisory representative of how that controller would prefer to solve the conflict. To achieve this, the approach built on the principal notion of using recordings of controller’s own solutions which where then disguised as auto- mated resolution advisories given later. As such, the approach undertaken here is not to develop an advanced CD&R algorithm, but rather to simulate decision aiding automation. If the solution suggested by the automation conforms with the problem- solving style of the controller, it is reasonable to expect that the match, as perceived by the controller, would benefit the controller’s acceptance of that solution. Philosophically, this thesis takes inspiration from the brilliant work of English mathematician Alan Turing, who many years ago proposed the ultimate test for ar- tificial intelligence, namely: if one can converse with a computer and not be able to distinguish its responses from those of a human, then that machine can truly be said to “think”. 38 In practical terms, inspiration was taken from a replay procedure car- ried out by Fuld et al. 39 for studying the impact of automation on error detection. In their study, automation performance was simulated by using unrecognizable replays of an operator’s own previously recorded performance. Results showed that opera- tors were more likely to attribute faults to automation than to themselves, when in fact, it was their own errors that they witnessed. Figure 1-2 depicts the method used for investigating strategic conformance (see also Appendix A). First, controllers participate in the prequel simulation , in which they play the same scenario(s) and manually solve the same (designed) conflict(s) multiple times. Conflict detection is supported by a short-term conflict advisory (safety-net) and a novel CD&R support tool. In the conformance design phase,