Recently many theories of the semantics and pragmatics of deontic modals have relied on analogies with decision theory. One particularly straightforward way to make this connection is to build the semantics around degree scales, where the degrees in question involve expected (moral, personal, etc.) value. This "Bayesian" semantics has a number of attractive features involving information-sensitivity, grammatical gradability, and solutions to logical puzzles which plague standard theories of deontic semantics. However, its viability has been questioned by several authors who argue that it is insufficiently expressive empirically or that it builds philosophically suspect assumptions into the semantic theory. These authors have advocated moving instead to a much weaker semantic theory which can express the predictions of the Bayesian theory among many others. In this paper I survey some of the arguments in favor of the Bayesian semantics and respond to objections, arguing that the empirical problems can be met, that the philosophical objections are unconvincing, and that, absent compelling empirical arguments, empirical data and methodological considerations favor a strongly predictive theory of deontic semantics over one with much greater expressivity.