More Problematic than the Newcomb Problems: Extraordinary Cases in Causal Decision Theory and Belief Revision

In this paper, I turn toward the types of cases regarding which I-CDT and K-CDT come apart. In particular, I present two novel examples, which, I argue, display I-CDT to be superior to K-CDT. The first, the “Chancy Dog Problem,” follows the form of a scenario that is explored in Rabinowitz [2009] and shows I-CDT to give what I argue to be the correct recommendation in a chancy universe, where K-CDT gives the incorrect one. I explain this difference as stemming from the fact that conditionalization shifts credences too much, adjusting beliefs as though the oracle provides more information that it actually does. Specifically, conditionalization leads one to revise one’s beliefs as though the oracle’s information tells you not only what will happen, but also what will happen were you to do something other than what you will actually do. Imaging by shifting credence directly to the nearest compatible world avoids this fault by making the minimal adjustment to one’s beliefs to maintain consistency. The other example, the “Faulty Signal Problem,” is structurally similar to a problematic scenario discovered by Collins [forthcoming], which reveals K-CDT to fail to provide a recommendation even in certain nonchancy universes, particularly those that exhibit what I refer to as “counterfactual asymmetries.” I explain why cases of counterfactual asymmetry present a particular difficulty for K-CDT, as a theory which utilizes conditionalization as a method of updating on the information provided by taking a certain action-proposition to be true. In both evidential and causal decision theory, one considers an action by revising one’s belief to suppose the action-proposition is true. The evidential decision theorist does this revision by hypothesizing that she receives, in an indicative fashion, news that the proposition is true. One who uses K-CDT does the same, but holds fixed certain causal beliefs. The Newcomb Problem and other such cases have shown, I believe, EDT to be wrong, but the problems I will present will show this second suggestion to be wrong as well. Instead, the correct form of decision theory requires an agent to subjectively place herself in different circumstances, even if such circumstances do not agree with facts she believes to be true. Deliberating in this subjunctive sense, as I discuss, requires imaging