Abstract

Major crisis incidents in the last couple of years show that there exist huge problems in the current practice of crisis management. It is a combination of failure in communication, failure in technology, failure in methodology, failure of management and finally failure of observation. It is therefore important that solutions are provided to prevent these kinds of incidents in the future. When a large incident should strike, the authorities are very interested in a system that can give possible hypotheses in an early stage of a crisis. Also there is a lot of room for improvement in the training of skills required for crisis management such as teamwork, coping with stress, filtering incoming messages and making the right command decisions based on the available information.
In search for a solution we tried to give an answer to the question to which extent it is possible to develop a Multi- Agent System and crisis environment supported by AI-reasoning to create a realistic simulation of a crisis and the corresponding crisis response by the authorities. By doing interviews with domain experts we came across the problems occurring in crisis management and requirements for a tool that crisis responders are waiting for. The proposed solution is called MACSIM, which stands for Multi-Agent Crisis Simulator, Interpreter and Monitor. It is a multi-component disaster simulation and detection system that, if fully developed, is a tool that can be used for realistic simulation of crisis situations. It is used for training purposes to get a quick assessment in the first few minutes after the unfolding of an event. The input of the system is based on two data sources, detectors and reports from people inside the concerned area. The system consists of a simulation component, a multi-agent communication layer and a set of agents acting inside a virtual world.
We tested the system with two example scenarios that implement everything that is currently possible with the system as far as showing simulation scenarios is concerned. Despite the fact that the currently implemented prototype is a proof of concept, the results of the two experiments are very promising because the system showed the expected behavior during the tests.