LingEQ Study: Most Cognitively Dense Paper in 90 Years of AI History Was Not About AI
Claude Shannon's 1948 communications paper recorded the highest linguistic entropy value, surpassing two of Alan Turing's landmark papers.
SILICON VALLEY, California – A study applying linguistic entropy measurement to 50 landmark artificial intelligence papers spanning 1936 to 2025 has produced a counterintuitive result: the text with the highest measured cognitive density was not written about AI.
Claude Shannon's 1948 paper A Mathematical Theory of Communication registers the highest Linguistic Entropy Quotient (LEQ) value in the entire corpus at 194 — surpassing Turing's 1936 On Computable Numbers (LEQ 193) and Turing's 1950 Turing Test paper (LEQ 189).
GPT-3, AlphaFold and DeepSeek-R1 cluster between 168 and 170, according to the study. The 1955 Dartmouth Proposal, which named the field of artificial intelligence, registers the lowest early-period value at 150.
Early AI texts created new conceptual objects and theoretical boundaries, the study says, while recent landmark work — however transformative — builds within established frameworks.
How it was measured
The study, 90 Years Engraved by Entropy: LEQ Analysis of 50 Landmark AI Papers, is the inaugural release of the LingEQ Curation Series. It was conducted using the LingEQ Engine, which uses multiple language models as analytical instruments, calibrated against an anchor corpus to produce reproducible linguistic entropy values.
Its measurements are derived entirely from the text itself, independent of authorship, publication venue or citation history. The theoretical framework is presented in The Scale of Language, forthcoming from Springer Nature.
A gap in scientific publishing
The research addresses what it describes as a structural gap in scientific publishing. arXiv now receives over 20,000 preprints each month, while papers can spend one to two years in peer review.
AI can generate increasingly polished texts, making originality and reliability harder to assess. Each submission is evaluated by two or three reviewers, introducing sampling bias and the limits of individual expertise.
Foundational work may disappear into an unread database, the study notes, while papers aligned with current trends can accumulate citations despite limited lasting value. "Peer review measures consensus, citation counts measure influence – neither measures what a text actually contains," it states.
The LingEQ Engine opened for global public beta on Aug 1, 2026. LingEQ Technologies describes itself as an independent language measurement platform quantifying the cognitive depth of written text in the age of AI, operating across Silicon Valley, Brisbane, Hong Kong and Shenzhen.
This article was produced with the assistance of artificial intelligence (AI), in accordance with our editorial policy.