Measurement Theory

Measurement theory is the study of how abstract quantities are mapped to observable indicators — the bridge between what we want to know and what we can actually record. It is not merely a branch of statistics but a foundational inquiry into the conditions under which a number can be said to represent a property of the world. Without measurement theory, scientific claims about 'intelligence,' 'capability,' or 'trust' are not empirical hypotheses; they are unexamined conventions dressed in numerical clothing.

The central problem of measurement theory is representation: under what conditions does a numerical structure preserve the relational structure of the empirical phenomenon being measured? The operationalist tradition, following Percy Bridgman, held that a concept is synonymous with the operations that measure it — a view that collapses meaning into procedure and leaves no room for criticizing the procedure itself. Modern measurement theory, particularly the representational theory associated with Krantz, Luce, Suppes, and Tversky, treats measurement as a homomorphism between an empirical relational structure and a numerical relational structure. This preserves the possibility that a measurement procedure can be wrong: the homomorphism may fail, and when it does, the numbers mislead.

The stakes are highest in domains where the target property is contested. AI capability claims routinely assume that benchmark performance maps to 'reasoning' or 'understanding,' but no measurement-theoretic argument establishes this homomorphism. The gap between what is measured and what is claimed is not a technical problem of better benchmarks. It is a conceptual problem of whether the target property has a structure that admits numerical representation at all. Measurement theory asks this question explicitly — which is why fields that ignore it tend to overclaim.