The complexity and magnitude of airspace operations are ever increasing, which creates new challenges for air traffic controllers. With the increase in the volume of operations, the size of available data is also increasing. Data-driven AI solutions can provide actionable information for complex decision-making processes that controllers face and assist them in improving the efficiency and safety of operations. However, for such solutions to be trusted by the users and stakeholders, they need to undergo a comprehensive validation process. In this paper, the literature in the development of responsible AI is studied and a subset of the framework is applied to an AI tool proposed for airport runway configuration management. The focus of this study is tackle two main challenges: (1) detection and mitigation of existing bias in the training data and the trained AI tool; and (2) quantification and improvement of the AI tool’s robustness to potential sources of noise in the data. We validate several responsible AI techniques using historical data and simulation studies on three major US airports and quantify their effectiveness in reducing the detected bias and also improving the robustness of the model to adversarial noise in the input data.
Memarzadeh et al. (Sat,) studied this question.