Competent inference and the (ir)rationality of level-splitting

In a 2012 paper, Ralph Wedgwood has argued for the view that competent inferences are always rational inferences. His theory of the significance of competent inferences jots down at a prominent debate in contemporary analytic epistemology, the debate around the rationality of level-splitting. Roughly, Wedgwood’s cases of competent inference have been pointed to as exemplifying a situation whereby level-splitting is rational. In this paper, I’ll argue against Wedgwood by challenging one of the assumptions underpinning his view, namely, the assumption that drawing an inference competently suffices to make it rational for an agent to endorse its conclusion in a way that is independent of the circumstances. As I’ll show, circumstances matter. For level-splitting to be rational in cases of competent inference in the way alluded, it would have to be the best response available, i.e., better than alternative responses the agent could have, but it is not at all clear that it is.