Evaluating Trust in an Uncertain and Multisource Environment

Recently, questions are emerging such as "how much" can one trust an Artificial Intelligence (AI) solution, how can the user interpret its outcomes, or how can the system explain the results provided? In the context of information fusion, the specific aspects of uncertainty representation and reasoning should be questioned the same way, as important components of the inference engine. This paper presents the impact from a subset of information quality dimensions on trust through the post-analysis of two serious games. Two scenarios are considered: Military Operations in Urban Terrain (MOUT) and Maritime Security Operations (MSO), in which individual players have to decide based on imperfect information from sources, alternatively human or artificial. The paper establishes a common framework for analyzing trust for situational awareness (or belief assessment) and decision making. On the one hand, in the CMRE Risk Game (MSO scenario), trust is measured through degrees of belief change, while in the GMU Game (MOUT scenario), trust is measured by the effective decision made. Results analyze and quantify the impact of source reputation and reliability, but the quality dimensions of confidence and specificity on trust. We finally provide next steps for future studies.