The Fluency-Authority Decorrelation

In the AI Story

The correlation's reliability across human history was not accidental. It reflected the material conditions of literate communication. Producing polished prose about a domain required sustained immersion in the domain — reading the literature, engaging with its problems, internalizing its characteristic vocabulary and rhetorical patterns. The process that produced fluency simultaneously produced authority, because both were products of the same extended engagement. The medical student who could write a competent case analysis had, in acquiring that fluency, developed the clinical knowledge that made the analysis authoritative. The correlation was structural: you could not, in general, acquire the surface without acquiring the depth.

Daston's historical research on scholarly Latin illuminates a precise precedent. For centuries, scholarly Latin served as both a medium of communication and a credentialing mechanism. A scholar who could write in proper Latin had, by definition, received the extended training that produced scholarly competence. The language was the credential — simultaneously a medium of expression and a signal of the education that made the expression trustworthy. When vernacular languages began to displace Latin as the medium of scholarly communication in the seventeenth and eighteenth centuries, the credentialing function was disrupted: texts in French or English could be produced by anyone who spoke those languages, regardless of formal training. New credentialing mechanisms — university degrees, professional societies, peer-reviewed journals — had to be constructed to replace the one the language shift had destroyed.

AI is producing an analogous disruption at greater scale and speed. The fluency that served as a natural credential can now be generated by a system that has not undergone the engagement the credential was supposed to certify. The credential has been decoupled from the process it used to certify. A text that reads like the product of deep expertise may be the product of statistical pattern-matching across a training corpus, and there is no feature of the text itself that allows the reader to distinguish between the two cases. The circularity is exact: the reader who needs the AI's summary lacks the expertise to evaluate it, and the reader who possesses the expertise does not need the summary.

The consequences extend beyond individual cases of misinformation. When fluency can be generated without the process that produces authority, the market's willingness to subsidize that process erodes. Why invest years in mastering a body of literature when a machine can produce a fluent summary in seconds? The question is not rhetorical. It is being answered, in admissions offices and hiring committees and editorial boards, in ways that will shape the intellectual infrastructure of the coming generation. The decorrelation is not merely an evaluative puzzle for individual readers; it is a structural pressure on the institutions that produce the very expertise the decorrelation reveals as newly precarious.

Origin

The specific formulation 'fluency-authority decorrelation' is this volume's synthesis of several streams of Daston's work: her analysis of scholarly Latin's credentialing function in Classical Probability in the Enlightenment, her account of confidence artifacts in Objectivity, and her genealogy of data in her 2022 book Rules. The phenomenon it names has been observed by many commentators on AI; what Daston's framework adds is the historical precision that locates it within a pattern of credentialing crises that have accompanied every major transformation in the media of learned communication.

Segal's Deleuze error episode provides the canonical contemporary illustration. The passage that Claude produced connecting flow to a misattributed Deleuzian concept was fluent, elegant, rhetorically confident — and substantively wrong. The fluency did not indicate the authority it had indicated in every prior encounter with similarly polished prose. The correlation had broken, invisibly, in the specific output where Segal happened to check.

Key Ideas

Historical correlation was structural. Fluency and authority were correlated because the material conditions of producing fluent prose also produced the expertise that made the prose authoritative.

Scholarly Latin as precedent. The displacement of Latin by vernaculars broke a prior credentialing mechanism; AI is producing an analogous but larger disruption.

The circularity problem. Users who consult AI for summaries lack the expertise to evaluate them; users with the expertise do not need the summaries — making the technology most unreliable exactly where it is most used.

Institutional consequences. When fluency can be generated without authority, the investment in producing authority becomes economically precarious — reshaping the institutions that have historically sustained expertise.

Invisible in the output. No surface feature of AI-generated text marks it as fluent-without-authority; the decorrelation is detectable only through external verification.

Debates & Critiques

One debate concerns whether the decorrelation is truly structural or whether improvements in AI reliability — through retrieval augmentation, chain-of-thought reasoning, or architectural refinement — will eventually restore the correlation. Optimists argue that the gap will close through engineering; skeptics respond that even if AI reliability increases in aggregate, the absence of a reliable signal distinguishing reliable from unreliable outputs means the decorrelation persists at the level that matters for evaluation. A related debate concerns whether new credentialing mechanisms — provenance tracking, institutional verification, expert certification — can be built quickly enough to compensate for the loss of the prior correlation.