artificial intelligence." The authors situate large language models (LLMs) within a longer history of non-trivial machines or technologies delegated partial agency to act semi-autonomously on behalf of humans. The authors argue that generative AI represents a distinct third phase of AI development, following logical-symbolic and data-driven approaches, characterized by abductive, probabilistic reasoning over vast corpora of published text rather than rule-based grammar or statistical classification alone.

Central to the paper is the claim that LLMs operate as an “anti-grammar.” Human meaning-making relies on generative grammar and multimodal transposition—continuous movement between instances and concepts, reference and agency, and across modes such as text, image, sound, and embodiment. By contrast, LLMs process meaning empirically through linear token-to-token relations encoded as weights, without hierarchical abstraction or intrinsic semantic understanding. Meaning remains latent within the model and becomes intelligible only when transposed back into human-readable text. Despite this limitation, generative AI exhibits powerful emergent capacities for hypothesis generation and reasoning at scale.

The authors further argue that the intelligence manifested in generative AI is neither fully artificial nor autonomous. It is deeply dependent on layers of human labor, linguistic conventions, fine-tuning, and socio-historical textual production. For this reason, they propose replacing the term artificial intelligence with cybersocial intelligence, emphasizing human–machine symbiosis rather than convergence. The paper concludes by highlighting the high stakes of this shift, including risks of concentrated ownership and epistemic inequality alongside possibilities for more equitable, dialogic, and multimodal forms of learning and knowledge production.

Cope, Bill, and Mary Kalantzis. 2025. "Generative Transpositions: The (Anti-)Grammar of Text-Semantic, CyberSocial Intelligence." Multimodality & Society 5 (4). https://doi.org/10.1177/2634979525138749

Generative Transpositions: The (Anti-)Grammar of Text-Semantic, CyberSocial Intelligence