University of Minnesota Press, ISBN 978-1517919320, English, 274 pages, 2025, USA

It should by now be universally understood that the entire technology underlying AI is based on the computation of language, as everything must have some form of textual description attached in order to be used in predictive processes. This appears to be the most deceptive aspect for many, who are seduced by the speed and plausibility of the outputs (the so-called ‘Eliza effect’), and fail to recognise the potentially disastrous fallibility of these semiotic systems in generating written outputs without explicit human intention. In the introduction to this book there is a key statement: ‘AI has accidentally divorced language from cognition’. ‘Meaning’ then occurs computationally ‘in the absence of intelligence’, and language is reduced to a ‘service’. Cultural and information theory are used to support this position and the call to return to a structuralist analysis of language. This would explain the substantial ambiguity between language’s direct reference to the world and the absence of any relationship between LLM text tokens and reality. Understanding the ideologies behind LLMs and how the algorithms function, can help us better understand how and why we use language. This is a dense book, but nonetheless an indispensable and ultimately stimulating one, which shifts our understanding of the AI conundrum to a higher level.