The Ramsey test and conditional semantics (Q1368751)

According to the Ramsey test, the conditional sentence \(A\to B\) (if \(A\) then \(B\)) is affirmed relative to a theory \(T\) if and only if \(B\) is entailed by \(T*A\), the outcome of revising \(T\) by \(A\). According to Stalnaker's truth condition for conditionals, the conditional sentence \(A\to B\) is true in a world \(\alpha\) if and only if \(B\) is true in the world \(f(A,\alpha)\), where \(f\) is a selection function that selects a world for every world \(\alpha\) and sentence \(A\). (Intuitively, it selects the \(A\)-world that is closest to \(\alpha\).) This paper is devoted to the connection between these two constructions. The author claims that if there is any such connection, then the Ramsey test should satisfy the following ``soundness'' condition: \vskip 2.5mm (1) From any theory that contains only true sentences, R[amsey] T[est] yields only true conditionals as conclusions. \vskip 2.5mm He shows that given (1) and plausible auxiliary conditions, the if-then operator \(\to\) referred to in the Ramsey test must coincide with material implication. Furthermore, he claims that if belief revision and world selection are at all ``related in any interesting way'', then they must be related in the following way: \vskip 2.5mm (2) \(B\in T*A\) iff \(B\) is valid in \(\{f(A,\alpha): \alpha\in [T]\}\), \vskip 2.5mm where \(| T|\) is the set of worlds in which \(T\) is true. It follows from (2) that \(*\) is not an AGM revision, but has instead the properties of an updating function. In order to avoid this problem, the author modifies Stalnaker's truth condition by relativizing it to a theory: \vskip 2.5mm (3) \(A\to_T B\) is true at world \(\alpha\) iff \(B\) is true at world \(f(T,A,\alpha)\). \vskip 2.5mm The author rightly observes that (3) exaggerates the context dependence of conditionals. He concludes that neither the binary nor the ternary selection function is convincing, so that we may have to give up either the Ramsey test or, at least, its connection with world selection. Although the general conclusion of the paper may be sound, it stands on uncertain ground since the author has only investigated some of the possible ways in which the Ramsey test and world selection may be connected. (1) is in fact quite implausible for the type of conditionals that the Ramsey test aims at explicating. For an example, suppose that I have no belief whatsoever about John's profession. My belief state (and the theory representing it) can then support the conditional ``If John goes home by taxi every day, then he is a rich man''. Since John is in fact a taxi-driver with low incomes, (1) requires that the Ramsey test does not support this conditional. It is not either obvious why (2) is the only interesting way in which revision \((*)\) and world selection \((f)\) can be connected. To consider just one alternative, let them both be based on a distance measure among worlds in the following way: \(T*A\) consists of the \(A\)-worlds that are closest to some \(T\)-world. \(f(A,\alpha)\) is the \(A\)-world that is closest to \(\alpha\). That (2) cannot be the correct account of belief revision was already realized by \textit{A. Grove} [``Two modellings for theory change'', J. Philos. Logic 17, 157-170 (1988; Zbl 0639.03025)], who consequently chose another construction. Finally, it is difficult to see how (3) can at all represent a truth condition. If truth is totally relativized to the agent's state to belief, in what sense is it still truth?