Similarity accounts of counterfactuals: A reality check1

To an unusual extent, philosophers agree that counterfactuals have truth conditions involving the most similar possible worlds where their antecedents are true, in the style of the celebrated and path-breaking Stalnaker/Lewis accounts. Roughly, these accounts say that the counterfactual if A were the case, C would be the case is true if and only if at the most similar A-worlds, C is true. I will argue that there are general structural problems with the appeals to both “the most” and “similar”. I will challenge any fixation on ‘the most __ worlds’, however we fill in the blank with a non-trivial ordering of worlds: in ignoring worlds that are later in the ordering, it adjudicates various implausibly specific counterfactuals to be true. I will then raise foundational problems for appealing to ‘similarity’—from consequents that are chancy, disjunctive antecedents, and unspecific antecedents more generally. I will also raise further problems for a number of specific proposals for understanding ‘similarity’. A recurring theme will be the tension that may arise between probability and similarity considerations. I will end by arguing for a paradigm shift, replacing ‘the most similar worlds’ approach with one based on conditional chances.