Artificial Intelligence Doesn't Exist

i.About the book

Artificial Intelligence Doesn't Exist argues that the term "artificial intelligence" has detached from the systems it now names. The phrase was coined in 1956 as the name of a research aspiration: the mechanization of intelligence itself. The systems we call AI today do not satisfy that aspiration. They are highly capable producers of language, and that capability is statistical and generative, not cognitive.

The paper proposes a more accurate description: artificial fluency. Large language models learn the statistical structure of public language and generate coherent, context-sensitive output. This is fluency, not the Dartmouth program of machine cognition. Treating fluency as evidence of understanding imports unsupported assumptions about judgment, reference, and responsibility — and human readers, formed by ordinary language to treat coherent speech as the expression of thought, supply the missing context themselves.

Across six chapters the book traces the misnaming from its historical source, describes what large language models actually do, distinguishes fluency from experience, examines the political economy that keeps the inflated term in circulation, documents the empirical harms in judgment-class domains, and makes the case for adopting "artificial fluency" as the accurate operating term.

ii.Why this book matters for KiTalent’s research

The paper is published by KiTalent Research. The distinction it draws is the same one that governs the firm's senior search practice: fluency is not judgment.

A candidate's CV, public profile, and AI-generated summary are fluent representations. They can be processed at scale. They are not the person. KiTalent's methodology treats representation as the input to judgment, never its substitute: mapping runs at maximum speed because availability is time-bound, while assessment runs at deliberate pace because judgment is what determines whether a senior placement holds. The paper gives that operating discipline its conceptual foundation — naming the difference between a system that produces language and a consultant who reads a person.

This book sits within a larger body of work. For the full map — which works are foundational, which are applied, and how the concepts connect — see A Guide to Alessio Montaruli's Research Corpus.