Distrust of Fluency

In The You On AI Encyclopedia

The heuristic that fluency indicates expertise was imperfect but generally reliable in a pre-AI information landscape. A human author who wrote fluently about a subject had, in most cases, spent considerable time understanding it, and the fluency was a byproduct of the understanding. A large language model that generates fluent text has processed statistical patterns in a training corpus, and its fluency is a byproduct of pattern-processing that may or may not correspond to genuine understanding.

You On AI documents a canonical failure of the old heuristic: a passage Claude produced connecting Csikszentmihalyi's concept of flow to a misattributed concept from Gilles Deleuze. The passage was rhetorically elegant, structurally coherent, and philosophically wrong. Only a reader who had independently engaged with Deleuze could detect the error — which is to say, only a reader who possessed the domain knowledge to evaluate substance independently of surface.

Mihaly Csikszentmihalyi

Blair's framework situates the challenge within the history of media-specific evaluative habits. Each new information technology has required the invention of new evaluative practices appropriate to the medium's distinctive surface features. The manuscript era valued provenance; the print era invented critical reading, source verification, and peer review. The AI era requires distrust of fluency and the associated practices of output interrogation.

The discipline has an affective dimension that mere cognitive technique cannot capture. Maintaining skepticism against confident, well-organized text requires sustained effort at a moment when the text is designed to reduce the sense that effort is necessary. The effort itself is a form of ascending friction — the relocation of difficulty from execution to evaluation that AI tools produce structurally.

Origin

The concept is an explicit extension of Francis Bacon's catalog of idols of the mind in the Novum Organum (1620). Bacon identified systematic sources of error in human cognition; an Idol of the Machine, as Blair's framework suggests, would name the specific conflation of fluency with truth that AI-generated content makes newly consequential.

Key Ideas

The heuristic is broken. Fluency no longer reliably indicates expertise, because fluency is now producible at industrial scale independently of understanding.

Surface-substance decoupling. AI produces uniform surface quality regardless of substantive accuracy; the decoupling is structural, not contingent.

Discipline, not cynicism. The goal is calibrated evaluation, not reflexive rejection; both excessive trust and excessive distrust are failures of the evaluative task.

Media-specific evaluative practices. Every medium requires its own evaluative discipline; the AI medium's discipline is still being invented.

Affective cost. Sustained distrust of fluency is cognitively expensive; it is the specific labor that ascending friction relocates the practitioner into.