Quantifying Rational Belief

Some criticisms that have been raised against the Cox approach to probability theory are addressed. Should we use a single real number to measure a degree of rational belief? Can beliefs be compared? Are the Cox axioms obvious? Are there counterexamples to Cox? Rather than justifying Cox’s choice of axioms we follow a different path and derive the sum and product rules of probability theory as the unique (up to regraduations) consistent representations of the Boolean AND and OR operations.

💡 Research Summary

The paper revisits the Cox formulation of probability theory, which treats probability as a quantitative measure of rational belief, and addresses several longstanding criticisms. First, it questions whether a single real number can adequately capture the degree of belief and whether different beliefs can be meaningfully compared. By assuming that belief ordering forms a complete preorder, the authors show that a monotonic real‑valued representation exists, but they acknowledge that human cognition often exhibits non‑linear, multi‑dimensional features that challenge this simplification.

Next, the authors scrutinize the “obviousness” of Cox’s three axioms: (i) preservation of belief ordering, (ii) a product rule for the logical AND, and (iii) a sum rule for the logical OR. They argue that these axioms are not merely intuitive truths but are grounded in two deeper mathematical requirements: operational consistency and invariance under re‑graduation (i.e., monotonic transformations of the belief scale). The product rule, for instance, holds exactly only when the two propositions are independent; dependence introduces deviations that can be absorbed by an appropriate re‑graduation function.

The paper then surveys purported counter‑examples to Cox’s system, such as non‑additive conjunctions, asymmetric disjunctions, and cases where an “impossible” event is assigned a non‑zero belief. In each case the authors demonstrate that a suitable monotonic transformation can map the problematic belief function back into a Cox‑compatible form, establishing that the Cox framework is essentially unique up to such re‑graduations.

The central technical contribution is a uniqueness theorem linking Boolean algebra’s AND and OR operations to a consistent numeric representation. Assuming the usual Boolean properties (associativity, commutativity, distributivity), the authors prove that any belief functions (F) (for AND) and (G) (for OR) that satisfy the consistency requirements must be expressible as

\