Leveraging explainable AI for enhanced decision making in humanitarian logistics: An adversarial coevolution (ACTION) framework
Abstract: This study examines the potential of AI-enabled wargames to enhance strategic decisionmaking in humanitarian assistance and disaster relief (HADR). We introduce an Adversarial CoevoluTION (ACTION) framework, which showcases AI’s capacity to evolve adaptable policies capable of responding to dynamic changes and adversarial actions in HADR wargame scenarios. The framework presented employs a grammar-based genetic programming algorithm to evolve intelligent and interpretable player policies. We apply the ACTION framework to a HADR wargame case study, commonly used by the Australian Defence Science and Technology Group for research purposes. The case study centres on a hypothetical disaster relief scenario in the fictional Joadia Islands, struck by a tsunami, necessitating the evacuation of dispersed civilians. Experimental results illustrate that the ACTION framework can evolve policies that adapt to environmental uncertainties and respond effectively to adversarial actions. This study offers evidence of the potential and practical application of AI-enabled technology in real-life humanitarian situations. Our findings suggest practical guidelines for humanitarian practitioners to enhance the efficiency and effectiveness of logistics planning for humanitarian aid, ultimately leading to improved outcomes in HADR scenarios.