Redundancy and Information

In The You On AI Field Guide

The framework begins with Shannon but Moles extends it into territory Shannon explicitly refused to enter. Shannon's 1948 theory treated information as a purely quantitative measure of uncertainty reduction, without reference to whether the message meant anything worth knowing. Moles argued that this limitation, while appropriate for engineers designing telephone systems, was inadequate for analyzing cultural messages, where the question of whether information is meaningful cannot be separated from the question of whether it is new.

The aesthetics of smoothness that Moles's framework diagnoses in contemporary AI-mediated culture is, in his terms, an aesthetics of maximal redundancy. The smooth interface, the frictionless experience, the seamless output — each systematically eliminates surprise. Jeff Koons's Balloon Dog, analyzed in You On AI, is Moles's concept rendered in stainless steel: perfectly smooth, perfectly predictable, perfectly without the accidents and imperfections that would carry information about its making.

The concern with AI-mediated production is that expanded channel capacity is being used to produce messages of increasing redundancy rather than increasing information content. A language model, absent specific instruction otherwise, tends toward the statistical center of its training distribution — toward the most probable continuation, which is by definition the most predictable, which is by definition the most redundant.

The temperature parameter in large language models is the explicit engineering control for this tradeoff. Low temperature produces highly predictable output (high redundancy); high temperature produces less predictable output (lower redundancy, but also potentially higher noise). The parameter makes Moles's axis operational. What remains human work is the judgment about where on the axis the output should sit.

Origin

Moles developed his treatment of redundancy in dialogue with Shannon, Weaver, and the French structuralist tradition. His particular contribution was to connect the mathematical notion of redundancy to long-standing aesthetic categories — banality, cliché, kitsch on one end; incoherence, obscurity, noise on the other — and to propose that aesthetic judgment is, at its core, calibration along this axis.

Key Ideas

Redundancy is the predictable proportion. A message's redundancy is measured by what a competent receiver could have guessed from context.

Information is the complement. What cannot be predicted carries information; what can be predicted carries none.

Both are necessary. Zero redundancy is incomprehensible noise; zero information is meaningless repetition. Meaning lives in the calibrated ratio.

AI defaults toward redundancy. Without specific intervention, language models produce the most probable continuation — the most predictable, the most redundant.

Kitsch is the aesthetic of maximal redundancy. Moles's critique of kitsch translates directly into a critique of a certain kind of AI-generated content.