Audun Stolpe, & Jo Hannay. (2021). On the Adaptive Delegation and Sequencing of Actions. In Anouck Adrot, Rob Grace, Kathleen Moore, & Christopher W. Zobel (Eds.), ISCRAM 2021 Conference Proceedings – 18th International Conference on Information Systems for Crisis Response and Management (pp. 28–39). Blacksburg, VA (USA): Virginia Tech.
Abstract: Information systems support to crisis response and management relies crucially on presenting actionable information in a manner that supports cognitive processes, and does not overwhelm them. We outline how AI Planning can be used viably to support the \emph{delegation and sequencing} of tasks. The idea is to use standard operating procedures as initial specifications of plans in terms of actors, actions and delegation rules. When expressed in the AI planning language \textit{Answer set Programming} (ASP), machine reasoning can be used in a \textit{pre-incident review} to display relevant delegation and sequencing inherent in a plan. % together with measures of goal achievement. The purpose of this is to uncover weaknesses in the initial plan and to optimize sequencing and delegation to increase the likelihood of achieving goals. Further, adaptive planning can be supported in \textit{during-incident reviews} by updating the current status, upon which ASP will then compute new alternatives. % and corresponding goal achievement measures. At this point, initial goals may no longer be viable and the explicit suggestion of prior sub-optimal goals now worth pursuing can be a game-changer under stress. The conceptual basis we lay out in terms of delegation and sequencing can be readily extended with further planning factors, such as resource requirements, role transfer and goal achievement.
|
Dashley K. Rouwendal van Schijndel, Jo E. Hannay, & Audun Stolpe. (2020). Simulation Vignette Generation from Answer Set Specifications. In Amanda Hughes, Fiona McNeill, & Christopher W. Zobel (Eds.), ISCRAM 2020 Conference Proceedings – 17th International Conference on Information Systems for Crisis Response and Management (pp. 110–121). Blacksburg, VA (USA): Virginia Tech.
Abstract: We investigate an approach that allows exercise managers to design simulations with an explicit focus on building skills, rather than having to focus on all the objects and interactions that a simulation must have. Exercise managers may design exercises at various levels of abstraction and always independently of how those sessions are implemented in simulations, while simulation components that implement the design are assembled and to some extent, automatically, behind the scenes. We outline (1) how Answer Set Programming can assist exercise managers in exercise planning and (2) how automated stage and content generation may be used to invoke appropriate simulation components to realize the design. For deliberate and recurrent training of decision-making skills, stages and content must vary to avoid familiarity (testing effects). We conclude by distilling a main research hypothesis that stipulates how (1) and (2) represent two modes of automated reasoning (so-called deductive versus abductive) and how that distinction clarifies the planning task.
|
Dashley Rouwendal van Schijndel, Audun Stolpe, & Jo Erskine Hannay. (2021). Toward an AI-based external scenario event controller for crisis response simulations. In Anouck Adrot, Rob Grace, Kathleen Moore, & Christopher W. Zobel (Eds.), ISCRAM 2021 Conference Proceedings – 18th International Conference on Information Systems for Crisis Response and Management (pp. 106–117). Blacksburg, VA (USA): Virginia Tech.
Abstract: There is a need for tool support for structured planning, execution and analysis of simulation-based training for crisisresponse and management. As a central component of an architecture for such tool support, we outline the design ofan AI-based scenario event controller. The event controller is a component that uses machine reasoning to computethe next state in a scenario, given the actions performed in the corresponding simulation (execution of the scenario).Scenarios are specified in Answer Set Programming, which is a logic programming language we use for automatedplanning of training scenarios. A plan encoding in ASP adds expressivity in scenario specification and enablesmachine reasoning. For exercise managers this gives AI-based tool support for before-action and during-actionreviews to optimize learning. In line with Modelling and Simulation as as Service, our approach externalizes eventcontrol from any particular simulation platform. The scenario, and its unfolding in terms of events, is externalizedas a service. This increases interoperability and enables scenarios to be designed and modified readily and rapidlyto adapt to new training requirements.
|
