Reports, Observations, and Belief Change

We consider belief change in a context where information comes from reports, and the reporting agents may not be honest. In order to capture this process, we introduce an extended class of epistemic states that includes a history of past reports received. We present a set of postulates that describe how new reports should be incorporated. The postulates describe a new kind of belief change operator, where reported information can either be believed or ignored. We then provide a representation result for these postulates, which characterizes report revision in terms of an underlying set of agents that are perceived to be honest. We then extend our framework by adding observations. In this framework, observations are understood to be highly reliable. As such, when an observation conflicts with a report, we must question the honesty of the agent that provided the report. We introduce a flexible framework where we can set a threshold for the number of false reports an agent can send before they are labelled dishonest. Fundamental results are provided, along with a discussion on key future problems to be addressed in trust and belief revision.