This paper employs epistemic decision theory to explore rational bridge principles between probabilistic beliefs and deductively cogent beliefs. I re-examine Hempel and Levi’s epistemic decision theories and generalize them by introducing a novel rationality norm for belief binarization. This norm posits that an agent ought to have binary beliefs that maximize expected utility in light of their credences. Our findings reveal that the proposed norm implies certain geometrical principles, namely convexity norms. Building upon this framework, I critically evaluate the Humean thesis in Leitgeb’s stability theory of belief and Lin-Kelly’s tracking theory. We establish the impossibility results, demonstrating that those theories violate the proposed norms and consequently fail to do the job of expected utility maximization. In contrast, we discover alternative approaches that align with all of the proposed norms, such as generating beliefs that minimize a Bregman divergence from credences. Our epistemic decision theory for belief binarization can be compared to Dorst’s accuracy argument for the Lockean thesis. We conclude that deductively cogent expected accuracy maximizers are neither Lockean nor Humean.

本文采用认识决策理论来探索概率信念和演绎有说服力的信念之间的理性桥梁原则。我重新审视了 Hempel 和 Levi 的认识决策理论，并通过引入一种新的信念二元化理性规范来推广它们。该规范假设代理应该具有二元信念，根据他们的可信度最大化预期效用。我们的研究结果表明，所提出的范数暗示了某些几何原理，即凸性范数。在这个框架的基础上，我批判性地评估了 Leitgeb 的信念稳定性理论和 Lin-Kelly 的跟踪理论中的 Humean 论点。我们建立了不可能性结果，证明这些理论违反了拟议的规范，因此无法完成预期的效用最大化的工作。相比之下，我们发现了符合所有拟议规范的替代方法，例如产生最小化 Bregman 与可信度的分歧的信念。我们关于信念二元化的认识决策理论可以与 Dorst 对洛克论文的准确性论证进行比较。我们得出结论，演绎有说服力的期望准确性最大化器既不是 Lockean 也不是 Humean。