What might be the case after a change in view (Q2499291): Difference between revisions

The author starts to point out that the constraints of Preservation and Reflectivity on belief revision are, on pain of triviality, inconsistent. In the paper he argues that the orthodox diagnosis, to opt for Preservation at the expense of Reflectivity, is wrong, although Preservation is the culprit. Let \(L^{+}\) be the smallest set containing classical propositional logic (CPL) such that, if \(\phi \in\) CPL, then \(\diamond \phi\), \(\neg \diamond \phi \in L^{+}\). Let us fix a space \(W\) of possible worlds. Any subset \(s \subseteq W\) is by definition a basic epistemic state. \(I\) is the set of all such basic epistemic states. Given a consequence relation \(\models\) \(\subseteq I \times L^{+}\), a state \(s\) is inconsistent, \(s = \bot\), iff there is some \(\phi \in L^{+}\) such that \(s \models \phi\) and \( s \models \neg \phi\). Otherwise \(s\) is consistent. Let \(\circ\) denote the revision function and \(B_{s} = \{\phi \in L^{+} ; s \models \phi \}\). PRESERVATION (P): For any \(s \in I\), \(\phi \in\) CPL: if \(s \not \models \neg \phi\) then \(B_{s} \subseteq B_{s \circ \phi}\). REFLECTIVITY: Your epistemic state commits you to \textit{it might be that} \(p\) iff it does not commit you to \(\neg p\). (And dually for must.) PERSISTENCE: Given a consequence relation \(\models\) \(\subseteq I \times L\) for a language \(L\). A formula \(\phi \in L\) is \textit{persistent with respect to} \(\models\) iff for any \(s, s' \in I\): if \(s \models \phi\) and \(s' \subseteq s\), then \(s' \models \phi\). The relation \(\models\) is \textit{persistent} iff all formulas in \(L\) are persistent with respect to it. The author argues that we should reject the doxastic conservative's (P) by rejecting the idea that rational epistemic commitment in the context of \textit{might} is persistent. Instead of a persistent consequence relation he introduces an independently motivated dynamic relation which yields us what we should want from epistemic models. Consider any \(s \in I\). A partial ordering \(\leq_{s}\) over \(W\) is an \textit{s-implausibility ordering} iff for any \(w \in s\): (1) for any \(w' \in W\), \(w \leq_{s} w'\), and (2) for any \(w' \not \in s\), \(w' \not \leq_{s} w\). Let min(\(\phi\), \(\leq_{s}\)) be the set of worlds \(w \in\) [[\(\phi\)]] such that no \(w' \in\) [[\(\phi\)]] is strictly less \(s\)-implausible than \(w\). A revision function \(\circ: I \times \) CPL \(\rightarrow\) \(I\) is \textit{broadly conditional} iff for any \(s\) and \(\phi \in\) CPL there is an \(s\)-implausibility ordering \(\leq_{s}\) such that \(s \circ \phi = \min(\phi, \leq_{s}\)). Let \(L^{\diamond}\) be the smallest set including CPL and such that if \(\phi, \psi \in L^{\diamond}\), then \(\neg \phi\), \(\phi \wedge \psi\), \(\diamond \phi\) \(\in L^{\diamond}\). The author proves the following theorem: Let \(M = \langle I, \models, \circ \rangle\) be a revision model for \(L^{\diamond}\), where \(\circ\) is a broadly conditional revision function, and consider any \(\phi \in\) CPL and \(s \in I\). (1) SUCCESS (S): \(s \circ \phi\) \(\models\) \(\phi\); (2) EASY REVISIONS (ER): If \(s \not \models \neg \phi\), then \(s \circ \phi = s \cap [[\phi]]\). (3) CONSISTENCY (C): If \([[\phi]] \neq \emptyset\), then \(s \circ \phi \neq \bot\). (4) NON TRIVIAL (NT): If \(W\) is non trivial, i.e., if there are two distinct possibilities, then \(M\) is non trivial.