An internal version of epistemic logic (Q965903)

The author discusses three different ways to model epistemic scenarios: The \textit{external approach}, where the modeller is not one of the agents. Depending on whether the modeller has perfect knowledge of the situation or not, this approach is either \textit{perfect} or \textit{imperfect}. The \textit{imperfect internal approach}, where the modeller is one of the agents, imposing a subjective point of view. The author notes that the imperfect internal approach, which could have applications in AI and cognitive psychology, has not been developed so far, and after reminding the reader on key notions in epistemic logic, he devotes the rest of the paper to proposing a possible formalism for that approach. The semantics is based on the notion of internal model, defined as a finite union of disjoint multi-agent possible worlds, each of which {\parindent8mm \begin{itemize}\item[(1)] is a Kripke model over a set of propositional interpretations with accessibility relations \(R_j\), one for each agent \(j\) under consideration, that satisfy three properties, namely, seriality, transitivity and Euclidicity; \item[(2)] has a selected structure \(w\) (the ``actual'' world) with the condition that \item[(2a)] for the particular agent \(Y\) that subjectively models the situation, \(w\) is the only possible world \(w'\) with \(R_Y(w,w')\) and \item[(2b)] for all \(j\neq Y\), no possible world \(w'\) satisfies \(R_j(w',w)\). \end{itemize}} By (2a), in each constituting multi-agent possible world with selected structure \(w\), \(Y\) considers \(w\) as the only possible world. Condition (2b) is meant to make the notion ``modular'' in the constituting multi-agent possible worlds. Let \(W\) denote the set of selected interpretations -- one for each constituting multi-agent possible world. The truth of a formula at a structure in the internal model is defined as usual, with the exception of the truth at \(w\in W\) of a formula of the form \(B_Y\phi\), meant to denote that agent \(Y\) believes \(\phi\) at \(w\), requesting that \(\phi\) be true at all members of \(W\). As a consequence, the model could equivalently be represented as a single Kripke frame built from all constituting multi-agent possible worlds by letting \(R_Y\) relate any member of \(W\) to any member of \(W\). Along those lines, the paper discusses some possible relationships between the external approach and the internal approach and, in particular, how internal models, one for each agent, can be extracted from a Kripke model that formalizes an external approach. Two dual notions of satisfiability and validity, a positive one and a negative one, are defined: negative satisfiability requires that some situation exists in which the formula is believed, whereas positive satisfiability requires that some situation exists in which the formula is not rejected. A complete set of inference rules is provided, and it is shown that the validity problem is decidable and PSPACE-complete. Besides illustrating the notions with a running example, the paper sketches how the proposed framework allows one to lift the AGM framework of belief revision to the multi-agent case.