Logical Traps in High-Level Knowledge and Information Fusion

In this paper, we are concerned with high-level information fusion, such as it is envisioned in JDL levels 4 and 5. More precisely, the focus is on decision and reasoning systems that must act in a rational and logical way from several knowledge and high-level information sources and databases. Whatever the selected knowledge and information representation languages in the sources are, an automatic reasoning system that is part of decision-support system must obey unquestionable rules and principles pertaining to logic. However, applying those rules that are initially defined for reasoning about a unique view of a situation, which is supposed to be fully described, can lead to serious and unexpected drawbacks in the context of the fusion of several different and partial views. We address three such issues. The first-one is related to the dramatic trivialising effects of logical inconsistency. The second one lies on the implicit transformation of necessary conditions into sufficient ones when fusion is operated. The last one is the implicit loss of more specific and precise information in favour of more general knowledge. Each of these problems is motivated by intuitive examples before practical solutions are discussed. 1