Towards the full information chain theory: question difficulty
A general problem of optimal information acquisition for its use in decision making problems is considered. This motivates the need for developing quantitative measures of information sources’ capabilities for supplying accurate information depending on the particular content of the latter. In this article, the notion of a real valued difficulty functional for questions identified with partitions of problem parameter space is introduced and the overall form of this functional is derived that satisfies a particular system of reasonable postulates. It is found that, in an isotropic case, the resulting difficulty functional depends on a single scalar function on the parameter space that can be interpreted – using parallels with classical thermodynamics – as a temperature-like quantity, with the question difficulty itself playing the role of thermal energy. Quantitative relationships between difficulty functionals of different questions are also explored.
💡 Research Summary
The paper addresses the fundamental problem of optimal information acquisition for decision‑making, emphasizing that the value of information lies not only in its quantity (as measured by classical entropy) but also in its quality—how accurate or reliable the information is for a specific decision context. To capture this nuance, the authors model a “question” as a partition of the problem’s parameter space Ω into mutually exclusive subsets {A_i}. For each question Q they introduce a real‑valued difficulty functional D(Q), intended to quantify the effort, cost, or resource consumption required from an information source to provide an answer of sufficient accuracy.
A set of intuitive postulates governs the form of D(Q). First, non‑negativity ensures that difficulty cannot be negative. Second, invariance under relabeling of the parameter space guarantees that two questions covering the same region of Ω have identical difficulty. Third, a “refinement” postulate states that if a question is subdivided into finer sub‑questions, the difficulty of each sub‑question must be a subset of the original difficulty. Fourth, an additivity (or quasi‑additivity) postulate requires that independent questions combine in a way that the total difficulty is the sum of the individual difficulties. These postulates collectively constrain the functional to a linear combination of a measure μ on Ω and a weight function w(ω) that reflects the local “hardness” of obtaining accurate information at point ω.
Mathematically, the derived form is
D(Q) = Σ_i μ(A_i)·w(ω_i),
where ω_i is a representative point of subset A_i. This expression captures both the size of the region addressed by the question (through μ) and the intrinsic difficulty of that region (through w). When the isotropy assumption is imposed—meaning the information source’s capability is uniform in all directions—the weight function collapses to a constant scalar T over the entire space. The authors interpret T as a temperature‑like quantity, drawing a direct analogy to classical thermodynamics: the difficulty D plays the role of internal energy, while T resembles temperature. In this picture, regions of higher “information temperature” demand more effort for the same level of accuracy, just as higher physical temperature requires more energy to change a system’s state.
The paper further explores relationships among difficulties of different questions. For overlapping questions Q₁ and Q₂, the shared region’s contribution to difficulty must not be double‑counted; the functional respects a sub‑additivity property akin to entropy’s behavior. Consequently, the difficulty of a combined question can be expressed as D(Q₁ ∪ Q₂) = D(Q₁) + D(Q₂) – D(Q₁ ∩ Q₂), providing a clear rule for aggregating information requests.
Practical implications are discussed. In sensor networks, each sensor’s measurement fidelity and energy consumption can be modeled by a local temperature T; designing query schedules that minimize total difficulty leads to more efficient data collection and faster decision cycles. In expert elicitation, an expert’s domain expertise can be mapped to a temperature field, allowing a decision analyst to predict which questions will be most costly in terms of cognitive load and to order queries optimally. The framework also suggests a systematic way to compare heterogeneous information sources by translating their performance characteristics into a common temperature scale.
Finally, the authors outline avenues for future work. Extending the theory to anisotropic settings—where w(ω) varies across Ω—would accommodate sources with direction‑dependent capabilities. Incorporating dynamic parameter spaces, where Ω evolves over time, would enable real‑time adaptation of question difficulty. Moreover, handling multiple interacting sources raises the prospect of defining joint temperature fields and cross‑source difficulty measures. Overall, the paper provides a rigorous, thermodynamically inspired foundation for quantifying question difficulty, opening new pathways for optimal information acquisition in complex decision‑making environments.