default A risk game formalization in support to maritime situation awareness automation: analysis of reasoning profiles


A risk game formalization in support to maritime situation awareness automation: analysis of reasoning profiles. Jousselme, Anne-Laure; Ben Abdallah, Nadia; Pichon, Fréderic. CMRE-FR-2016-011. December 2016.

This document follows on previous work on the Risk Game, a general methodology developed at CMRE to elicit experts knowledge and know-how, and their ability to deal with information of different nature, to consider the information quality and to reason about concurrent events. It is a contrived technique aimed at capturing data expressing human reasoning features while performing a specific task of maritime situation assessment. This report is mainly dedicated to the mathematical formalisation of the risk game and the underlying maritime situation assessment problem. We propose an overarching framework for maritime situation assessment which clearly distinguishes between the elements of (1) context, (2) source quality and (3) situation, and considers their influence links. The model has three layers: variable layer, valuation layer and information quality layer. We adopt the algebraic framework of Valuation Algebras allowing a great flexibility in uncertainty representation and reasoning. In order to cover a wide range of uncertainty representations, and further study the impact of modelisation choices, we frame the problem into the belief function theory as it naturally extends both classical sets (hence classical logic) and probability theory. Furthermore, we use a graphical model which, besides efficient computation, allows a clear exposition of the problem and solution for a hopefully better understanding for the user. The analysis performed illustrates comparative reasoning schemes between players and automated reasoners. By testing a few different schemes, we have been able to identify players whose reasoning was consistent with evidential reasoning, or who answered to a threat assessment problem rather than an identification problem. We have also been able to highlight the impact of context and source quality. Rather than suggesting a normative way of reasoning, this study aims at characterising personal reasoning attitudes of experts. Knowing in advance the consequences of a particular reasoning scheme could help prevent some reasoning biases.